change

A history of computing & internet security

About this history

We used to
trust that.

You could borrow a floppy disk and boot a PC from it. Years later, opening an invoice might mean clicking “Enable Content.” This is a history of things we did at the computer without giving them much thought, and how software changed once the risks became clear.

Explore the history

A familiar thing, changed

2021
A site with Flash content
Content blocked
Flash Player

Adobe blocked Flash content from running in Flash Player on January 12, 2021.

Read this part of the history

A simplified comparison; each entry explains the scope.

What changed, and why

46 shifts • 1980s to 2026

Entries are grouped by a documented turning point. Date ranges show the longer story; they don’t mean everyone changed at once.

Showing all 46 entries

1980s–1990s

When trust travelled on a disk

A floppy disk could carry a virus from one PC to the next. On the early internet, a mail server might forward messages for strangers, and a remote login could expose a password to anyone listening on the network.

1986–1992Software
Added defense

The Brain virus could spread from a floppy disk when an IBM PC booted up. ESET dates its first DOS detector to 1987 and its first retail sale to 1990 in Austria. By then, NIST was recommending virus checks before new software went into use; its 1992 guide helped organizations choose from a range of available tools.

The change, in detail
Before
IBM PC users exchanged floppy disks and software without a well-established scanner category or routine screening practice documented in federal guidance.
What put it under pressure
Boot-sector viruses such as Brain could spread when a computer booted from an infected diskette. As viruses multiplied, organizations needed ways to detect and contain infected software before sharing it.
What changed
NIST's 1989 management guide recommended antivirus software to test new software before wider release and advised controls on disk use. By December 1992, NIST described a variety of commercial and emerging antivirus tools and published criteria for choosing among them. One vendor's retrospective dates its first DOS detector to 1987 and its first retail sale to Austria in 1990.
What followed
Antivirus scanners became an identifiable product class and part of federal guidance for managing PC software. The evidence establishes availability and recommended practice, not widespread installation or compliance.
Where this applies
IBM PC and other personal computer environments, especially organizations selecting antivirus tools. Brain and ESET illustrate specific cases, not all virus families or vendors.
Origins and adoption
NIST dates Brain's appearance to 1986. ESET retrospectively dates its detector to 1987 and retail sale in Austria to 1990. NIST guidance appeared in August 1989 and its dedicated tool-selection guide in December 1992. The 1992 NIST guide states that a variety of antivirus tools were available, while the 1989 and 1992 documents recommend or help choose use. Neither gives market share or installation rates. ESET's evidence concerns one vendor and a regional first sale.

Dates in the record

  1. 1986The Brain IBM PC boot-sector virus appeared; NIST later describes floppy exchange as a PC-virus propagation route.
  2. 1987ESET's founders retrospectively date an MS-DOS virus-detection program to this year.
  3. 1989-08NIST published a management guide recommending antivirus checks and controls for software and disks.
  4. 1990ESET retrospectively dates its first retail sale, in Austria under the STOPVIR name, to this year.
  5. 1992-12NIST published a dedicated guide for selecting among antivirus tools and techniques.

Sources and what they establish

1988Networks
Added defense

When the Morris worm spread in November 1988, it disrupted the same research network people needed to coordinate their response. DARPA asked Carnegie Mellon’s Software Engineering Institute to establish CERT/CC, giving responders a standing point of contact to share information and coordinate with vendors.

The change, in detail
Before
Network administrators and researchers relied on informal contacts and disconnected local responses when a new threat spread across the ARPANET.
What put it under pressure
The worm replicated across networked Unix systems, consumed system resources, and made some communication paths unavailable just as responders needed them.
What changed
DARPA commissioned SEI to create a central coordination capability. CERT/CC was established in 1988 and distributed incident information and response guidance.
What followed
Incident response gained a standing coordination point. CERT/CC later helped other organizations form response teams, although the scale of the 1988 outbreak remains disputed across sources.
Where this applies
CERT/CC initially served research and government Internet users. It later helped national and organizational response teams.
Origins and adoption
The incident began November 2, 1988. DARPA's decision followed a post-incident review. CMU/SEI sources give differing month labels for the center's establishment, so the year 1988 is used here. The initial impact was significant within the small connected research Internet, but sources vary between approximately 4,000 affected systems and estimates of 5-10% of roughly 60,000-80,000 hosts. The later response-team network grew over decades.

Dates in the record

  1. 1988-11-02The Morris worm began spreading.
  2. 1988DARPA established the CERT Coordination Center at SEI in response to the incident.

Sources and what they establish

Early 1990s–2002Networks
Added defense

As organizations connected their internal networks to the internet, firewalls let them control which traffic crossed that boundary. By 1992, a USENIX paper described gateways already in use at AT&T and Digital Equipment Corporation. The IETF documented expected firewall behavior in 2000, and NIST published a deployment guide in 2002. A firewall at the edge left traffic between internal machines to other controls.

The change, in detail
Before
Connected organizations had to decide which network services could cross between internal systems and less trusted external networks.
What put it under pressure
Internet connectivity exposed internal services to remote access and attack, while unrestricted inbound traffic made every service harder to protect individually.
What changed
Network administrators installed packet filters, proxies, and gateway firewalls at network boundaries, with rules defining which traffic could pass.
What followed
Perimeter filtering became an established layer of organizational security. It did not detect every attack and could not stop traffic that moved only between internal hosts.
Where this applies
Organizational network security, with different firewall technologies and policies across organizations. Evidence establishes an established field by 2000-2002, not a universal installation date.
Origins and adoption
AT&T and DEC were among the organizations already operating gateways documented by 1992. The guidance establishes firewalls as a standard organizational concern but does not provide a measured share of organizations deploying them.

Dates in the record

  1. 1992A USENIX paper documents application gateways in use at AT&T, DEC, and other organizations.
  2. 2000-10RFC 2979 documented firewall behavior and interoperability expectations.
  3. 2002-01NIST published SP 800-41 guidance on firewall selection, deployment, and policy.
  4. 2009NIST described the limits of perimeter-only placement and the use of internal controls.

Sources and what they establish

1995–1999Networks
Added defense

After a password-sniffing incident at his university, Tatu Ylönen developed Secure Shell in 1995. SSH let people log in and copy files over an untrusted network without sending the session in plain text, and gave them a way to check which server they were connecting to. OpenBSD included the free OpenSSH implementation in 1999, years before the IETF standardized SSH’s architecture in 2006.

The change, in detail
Before
Telnet, rlogin, and related tools often carried login credentials and session data in clear text across shared networks.
What put it under pressure
A packet sniffer on a university network captured passwords, demonstrating that attackers who could observe traffic could reuse those credentials.
What changed
Ylönen wrote and released SSH in 1995. OpenBSD incorporated the open-source OpenSSH implementation in 1999; the IETF later published SSH as a standards-track protocol in 2006.
What followed
Encrypted remote administration became a common cross-platform practice. SSH did not immediately remove clear-text protocols, and host-key verification remained necessary to resist impersonation.
Where this applies
Administrators, researchers, and later general users of Unix-like systems and other platforms. The initial incident and implementation history are creator-reported.
Origins and adoption
Ylönen dates the first SSH implementation to 1995 and links its origin to password sniffing. OpenBSD's OpenSSH release history begins in 1999; RFC 4251 was published in January 2006. SSH became broadly used, but the cited standards and project histories do not quantify adoption or prove that Telnet and rlogin disappeared.

Dates in the record

  1. 1995Ylönen developed and distributed the first SSH implementation.
  2. 1999OpenBSD integrated OpenSSH.
  3. 2006-01The IETF published SSH architecture as RFC 4251.

Sources and what they establish

1996–2001Networks
Added defense

In 1996, Lawrence Berkeley National Laboratory deployed Bro to watch live network traffic for signs of intrusion. Bro could flag suspicious connections for investigation. By 2001, NIST was explaining how organizations could add intrusion detection to both hosts and networks.

The change, in detail
Before
Many organizations relied on access controls and perimeter devices that could not reveal all suspicious activity within systems or across allowed network paths.
What put it under pressure
Network attacks increased in number and severity, creating demand for ways to monitor events and identify possible intrusions.
What changed
Berkeley Lab deployed live monitoring in 1996. NIST described host and network intrusion detection in 2001.
What followed
Operators at Berkeley Lab could investigate alerts from passive monitoring. Later NIST guidance helped organizations evaluate IDS.
Where this applies
Organizational information systems and networks, as described by NIST. The guide's claim that IDS was necessary for most organizations reflects NIST's assessment, not measured deployment prevalence.
Origins and adoption
Berkeley Lab began operational deployment of Bro in 1996. Bro establishes operational use at Berkeley Lab. NIST guidance gives no industry deployment rate.

Dates in the record

  1. 1996Berkeley Lab begins operational deployment of Bro.
  2. 1998USENIX publishes the Bro research.
  3. 2001-11NIST publishes its guide to host and network intrusion detection.

Sources and what they establish

1998–1999Networks
Changed default

An open mail relay would pass along messages even when neither the sender nor the recipient belonged to its own network. Spammers turned that courtesy into free delivery infrastructure, leaving administrators with full queues and a damaged reputation. Sendmail 8.9 disabled third-party relaying by default in 1998; the following year, Internet best-practice guidance made relay controls explicit.

The change, in detail
Before
Forward mail for unrelated senders and recipients.
What put it under pressure
Spam operators borrowed other people’s mail servers.
What changed
Refuse unauthorized third-party relaying.
What followed
Sending through a server required an authorized route.
Where this applies
Sendmail default and Internet operational guidance; not an overnight change to every server.
Origins and adoption
SMTP relay architecture predates the commercial web. RFC 2505 describes widespread abuse and operational costs, not a census.

Dates in the record

  1. 1998-05Sendmail 8.9 default changed
  2. 1999-02RFC 2505 published

Sources and what they establish

Back to filters
2000–2009

Windows gets new defaults

A Word attachment could mail copies of itself, and a worm could find a Windows PC before anyone had turned on its firewall. Microsoft began putting out security bulletins on a schedule, while browsers started putting limits on what a web page could reach.

1999–2001Software
Added defense

Melissa spread through Word macros and Outlook address books. After the ILOVEYOU outbreak, Microsoft released an update for Outlook 98 and 2000 in June 2000. It blocked access to certain executable attachments and prompted users when another program tried to send mail or read the address book. Outlook 2002 incorporated the controls.

The change, in detail
Before
Outlook users could open many executable attachments with few built-in warnings or restrictions, while macro-enabled documents could automate actions in Office applications.
What put it under pressure
Melissa's macro and Outlook automation spread infected Word documents through address books. ILOVEYOU soon used email attachments to spread a script-based worm.
What changed
The June 2000 update blocked access to specified executable attachments and added prompts for programmatic sending and address book access.
What followed
The controls shipped as an update and were incorporated into Outlook 2002.
Where this applies
Microsoft Outlook installations receiving the update. Other mail clients and server-side filtering products had separate controls and schedules.
Origins and adoption
Melissa was first detected in March 1999. Microsoft announced the Outlook update in May 2000 and released it in June. Broad potential within Outlook users, but uptake varied by release, service pack, and administrator configuration. No global installation or effectiveness measure was established here.

Dates in the record

  1. 1999-03-26Melissa began spreading through macro-enabled Word documents and Outlook.
  2. 2000-05-04The ILOVEYOU worm spread through a script attachment.
  3. 2000-05-15Microsoft announced the Outlook Email Security Update.
  4. 2000-06-08Microsoft releases the Outlook Email Security Update for download.
  5. 2001Outlook 2002 incorporates the protections.

Sources and what they establish

1998–2005Software
Retired or blocked

In July 1998, the Electronic Frontier Foundation’s Deep Crack machine recovered a 56-bit DES key in about 56 hours. The following January, Deep Crack and distributed.net cut that to 22 hours and 15 minutes. NIST published AES in November 2001, and the government withdrew DES as a federal standard in 2005. Some federal applications could still use Triple DES.

The change, in detail
Before
DES remained a federal encryption standard, even as computing power made its 56-bit key increasingly vulnerable to exhaustive search.
What put it under pressure
EFF's DES Cracker demonstrated practical key recovery in less than three days in July 1998; EFF and Distributed.net completed a later challenge in 22 hours 15 minutes in January 1999, while NIST had already begun a public competition to select a replacement.
What changed
NIST selected and published AES as FIPS 197 in November 2001. The Secretary of Commerce withdrew DES-related FIPS standards in May 2005 and encouraged agencies to use AES or, for some applications, Triple DES.
What followed
DES ceased to be an approved federal encryption algorithm. AES became the federal successor, while some uses continued under Triple DES and private-sector transitions varied.
Where this applies
Federal information processing standards and organizations. The change did not automatically reconfigure all commercial products or deployed systems.
Origins and adoption
EFF's public demonstration occurred in January 1999. AES was published November 26, 2001 and became effective May 26, 2002. DES was withdrawn May 19, 2005. Clear formal shift within federal standards. The sources establish recommendation and withdrawal, not the completion date for every agency or commercial migration.

Dates in the record

  1. 1998-07-17EFF's DES Cracker won DES Challenge II-2 in less than three days.
  2. 1999-01-19EFF and Distributed.net won DES Challenge III in 22 hours 15 minutes.
  3. 2001-11-26NIST published AES as FIPS 197.
  4. 2002-05-26FIPS 197 became effective.
  5. 2005-05-19NIST withdrew DES-related FIPS standards.

Sources and what they establish

2003Software
Added defense

Code Red and Nimda spread through vulnerabilities that already had patches. Administrators also told Microsoft they needed predictable release dates. In October 2003, Microsoft moved security bulletins to a monthly schedule so customers could plan testing and installation.

The change, in detail
Before
Microsoft released security fixes when they were ready, leaving administrators to respond to releases on an unpredictable schedule.
What put it under pressure
Code Red and Nimda exploited flaws despite patches having been available, showing the operational gap between vendor release and administrator deployment. Microsoft also received customer requests for predictable timing.
What changed
Microsoft announced a monthly security bulletin cadence on October 1, 2003, alongside tools intended to help customers scan and deploy updates.
What followed
Administrators gained a regular planning date for Microsoft security releases. A predictable release schedule did not itself ensure timely installation or patch compliance.
Where this applies
Microsoft products and customers managing those products. Other vendors and organizations used different update schedules.
Origins and adoption
Microsoft announced the monthly schedule on October 1, 2003. The first monthly security update followed in October. The later label Patch Tuesday describes a Microsoft release schedule, not an industry standard. The cadence became a durable Microsoft practice, but the sources do not show how many customers installed fixes or how quickly they did so.

Dates in the record

  1. 2001-07Code Red spread through a vulnerability for which a patch had already been released.
  2. 2001-09Nimda exploited multiple previously patched vulnerabilities.
  3. 2003-10-01Microsoft announced monthly security bulletins.

Sources and what they establish

2004Software
Added defense

Buffer-overflow exploits could place instructions in memory and then run them. In 2004, Windows XP Service Pack 2 added hardware-enforced Data Execution Prevention on compatible PCs. The default covered essential Windows programs and services, and administrators could enable broader coverage.

The change, in detail
Before
Many systems allowed code to execute from memory regions used to hold ordinary data, leaving some buffer-overflow attacks able to run injected instructions.
What put it under pressure
Buffer overflows were a common exploit vector, and attackers used them to place and execute code in writable memory.
What changed
Windows XP SP2 added hardware-enforced Data Execution Prevention on compatible hardware and used an OptIn default for essential Windows programs and services.
What followed
The operating system could prevent execution from some data pages, raising the difficulty of certain exploits. DEP depended on processor, BIOS, Windows version, and configuration and did not remove memory-corruption vulnerabilities.
Where this applies
Compatible processors running supported Windows versions. Defaults differed between 32-bit and 64-bit programs and across policy settings.
Origins and adoption
Windows XP Service Pack 2 introduced hardware-enforced DEP support in 2004. Hardware enforcement required processor and firmware support. XP’s default covered essential Windows programs and services.

Dates in the record

  1. 2004-08Windows XP Service Pack 2 ships with hardware-enforced DEP support on compatible systems.

Sources and what they establish

2001–2010Networks
Still changing

Researchers showed in 2001 that captured Wi-Fi traffic could reveal a WEP key. WPA offered an interim replacement, and WPA2 followed the stronger IEEE standard in 2004. The payment card industry set a deadline for wireless networks in scope: existing installations had to stop using WEP after June 30, 2010.

The change, in detail
Before
Wi-Fi networks commonly relied on WEP, whose key scheduling and reuse properties weakened the protection it was meant to provide.
What put it under pressure
Researchers published practical cryptanalysis of WEP's use of RC4 in 2001, demonstrating that captured wireless traffic could reveal key material.
What changed
The Wi-Fi Alliance offered WPA as an interim option and WPA2 for IEEE 802.11i. PCI DSS set scoped WEP retirement deadlines.
What followed
Network owners gained replacement options. PCI DSS prohibited WEP on existing wireless implementations in its cardholder data scope after June 30, 2010.
Where this applies
IEEE 802.11 wireless LAN standards and Wi-Fi Alliance certified products. This transition does not describe every wireless network or imply that WEP was quickly removed everywhere.
Origins and adoption
The cited WEP cryptanalysis appeared in August 2001. WPA was introduced in October 2002. IEEE approved 802.11i on June 24, 2004 and published it July 24, 2004. The deadlines applied to wireless networks that transmitted cardholder data or connected to the cardholder data environment. Consumer adoption varied.

Dates in the record

  1. 2001-08Researchers demonstrate a practical WEP key recovery attack.
  2. 2002-10The Wi-Fi Alliance publishes interim WPA.
  3. 2004IEEE publishes 802.11i and WPA2 follows.
  4. 2009-03-31PCI DSS deadline for new wireless implementations in scope to avoid WEP.
  5. 2010-06-30PCI DSS deadline for existing wireless implementations in scope to stop using WEP.

Sources and what they establish

2004Software
Changed default

Earlier releases of Windows XP came with a firewall and left it off. After network worms such as Blaster and Sasser exploited exposed services, Service Pack 2 turned on the built-in firewall by default in August 2004 when it found no third-party firewall. Administrators could still set exceptions or change the policy.

The change, in detail
Before
Windows XP included Internet Connection Firewall, but the default configuration left it disabled.
What put it under pressure
Blaster and Sasser spread by exploiting network-accessible Windows services. Microsoft said the firewall default would stop these worms before they reached the PC.
What changed
Windows XP SP2 replaced the earlier firewall with Windows Firewall and enabled it by default when another firewall was not detected.
What followed
Many Windows XP systems received inbound traffic filtering without users having to turn it on manually. Existing third-party firewalls, enterprise policies, service exceptions, and configuration differences limited uniformity.
Where this applies
Windows XP SP2 installations. It does not describe all Windows versions or every home and enterprise configuration.
Origins and adoption
Microsoft released SP2 to manufacturing on August 6, 2004; consumer distribution followed later in August. The firewall default changed with SP2, not with Windows XP's original release. XP had a very large installed base, but Microsoft and NIST describe the feature and defaults rather than reporting a measured fraction of systems protected by the firewall.

Dates in the record

  1. 2003-08Blaster spread through exposed Windows RPC services.
  2. 2004-07Microsoft publicly described the SP2 firewall default as a response to Blaster and Sasser.
  3. 2004-08-06Windows XP SP2 was released to manufacturing.
  4. 2004-08Windows XP SP2 reached consumers.

Sources and what they establish

2006–2007Software
Changed default

With Windows Vista’s User Account Control, even an administrator did everyday work with reduced privileges. A program that needed more access could bring up a prompt, and some older apps could keep working through compatibility features without getting full administrator rights. Users could suppress the prompts, and the protection depended in part on what they approved.

The change, in detail
Before
Many Windows users and applications operated with administrator privileges for routine tasks.
What put it under pressure
Malware running as the logged-in user could make persistent system-wide changes, while legacy software often assumed it could write to protected locations.
What changed
Vista introduced split tokens, consent or credential prompts, and compatibility features intended to make standard-user operation more practical.
What followed
UAC brought elevation into ordinary desktop use and pushed developers toward privilege-aware software, while prompt fatigue and compatibility issues led to criticism and configuration changes.
Where this applies
Windows Vista and later Windows versions; exact prompt behavior depends on account type and policy.
Origins and adoption
Microsoft's Windows Vista engineering program, building on least-privilege operating-system design. Shipped with Vista and persisted in later Windows releases. The sources establish availability and design, not the share of users running standard accounts or leaving prompts enabled.

Dates in the record

  1. 2006-11-30Windows Vista became available to business customers.
  2. 2007-01-30Windows Vista general availability began.

Sources and what they establish

2008Web
Added defense

From the start, Chrome rendered pages in a restricted process that had to go through the browser to get at files and the network. A bug in the code that drew a page could still give an attacker control of that process, but the sandbox was designed to limit what the attacker could reach from there.

The change, in detail
Before
A browser code-execution flaw could give the exploited component broad access to a user's files and operating-system resources.
What put it under pressure
Web content exposed large parsing and scripting engines to untrusted input, making complete prevention of code-execution flaws unrealistic.
What changed
Chromium separated renderer and browser processes and restricted renderer resource access through a sandbox policy.
What followed
The browser gained a containment layer after code execution. Later sandbox escapes and platform-specific limitations show why it remained one layer rather than a full boundary guarantee.
Where this applies
Chrome/Chromium; sandbox capabilities and restrictions vary by operating system and browser version.
Origins and adoption
Chromium project, publicly described alongside the 2008 browser launch. Adopted as a core Chrome design and subsequently common across major browser architectures; no market adoption statistic included here.

Dates in the record

  1. 2008-09-02Google announced the Chrome beta and its security architecture.
  2. 2008-09-10Chromium published a technical explanation of its sandbox model.

Sources and what they establish

2008–2009Networks
Added defense

A forged DNS reply could send people to the wrong server even while their browser showed the address they expected. After researchers showed how easily attackers could poison a DNS cache, vendors issued coordinated patches in 2008 that made requests harder to guess, in part by varying which port they came from. An IETF specification the next year spelled out stricter checks on replies; cryptographic authentication was a separate defense.

The change, in detail
Before
Some resolvers used predictable source ports and limited response matching, giving attackers a better chance to inject forged DNS answers into caches.
What put it under pressure
Researchers demonstrated that attackers could exploit the limited entropy of DNS query identifiers to poison caches and redirect users.
What changed
RFC 5452 standardized strict query-response matching and required unpredictable source ports as additional entropy.
What followed
Resolver implementations gained a widely applicable defense against off-path forgery. It raises attacker cost but does not authenticate DNS data as DNSSEC does.
Where this applies
Recursive DNS resolver software and operations; it does not guarantee that each deployed resolver follows the standard.
Origins and adoption
IETF DNS Extensions Working Group, responding to DNS cache poisoning research and attacks. CERT lists affected implementations and vendor fixes; installed patch coverage is not quantified.

Dates in the record

  1. 2008-07-08Coordinated advisory and vendor patch release.
  2. 2009-01RFC 5452 published resolver forgery-resistance requirements.

Sources and what they establish

2009–2011Devices
Changed default

Malware such as Conficker spread through removable drives. Windows 7 removed the drive’s AutoRun program from the AutoPlay menu for USB mass storage. Microsoft later offered the change through automatic updates for Windows XP and Vista. CDs and DVDs kept their AutoRun behavior.

The change, in detail
Before
Windows could offer or launch programs described by autorun.inf when removable media was inserted, creating a path for malware propagation and misleading prompts.
What put it under pressure
Microsoft reported increased malware abuse of AutoRun through USB media in early 2009, with Conficker as a prominent example.
What changed
Windows 7 removed the AutoRun-supplied task from AutoPlay for non-optical removable media.
What followed
USB insertion no longer triggered the same automatic launch path by default, though optical media and some devices identifying as CD/DVD remained exceptions.
Where this applies
Windows 7 and later XP/Vista updates; non-optical removable media. Not a universal ban on AutoPlay or CD/DVD AutoRun.
Origins and adoption
Microsoft Windows engineering and MSRC in response to changing removable-media malware behavior. Became a Windows default with Windows 7 and was backported. Sources document intended behavior, not measured infection reduction.

Dates in the record

  1. 2009-04-28Microsoft announced the AutoRun change for Windows 7 Release Candidate and planned XP/Vista updates.
  2. 2009-10-22Windows 7 reached general availability.
  3. 2011-02Microsoft offers KB971029 through automatic updates for older Windows versions.

Sources and what they establish

Back to filters
2010–2019

The browser changed the rules

Installing a plug-in was once a normal part of visiting a site. As Flash and Java applets lost their place in the browser, HTTPS went from something you checked for at the checkout to something Chrome called out when it was missing.

2011–2012Web
Added defense

With HSTS, a site could tell a browser to use HTTPS on future visits and to refuse connections with certificate errors. The browser learned the rule over a secure connection, so an unprotected first visit could still be intercepted. Preload lists gave browsers a way to know the rule before visiting included sites.

The change, in detail
Before
A user entering a site's HTTP address or following an insecure link could be kept on plaintext HTTP by an active network attacker.
What put it under pressure
SSL stripping and mixed transport allowed attackers to downgrade some users before they reached the site's secure endpoint.
What changed
Browsers enforced site HTTPS policies, and Chromium documented preload support in 2011. The IETF standardized HSTS in 2012.
What followed
Returning users and preloaded sites can avoid insecure HTTP negotiation. Initial visits without preload and misconfigured subdomains remain limitations.
Where this applies
Participating browsers and sites that deploy HSTS correctly; preload only covers domains accepted into a browser's list.
Origins and adoption
Chromium documented preload support in 2011. The IETF published RFC 6797 in 2012. Implemented across major browsers, but RFC publication and support do not measure site deployment. Preload adoption varies by domain and browser.

Dates in the record

  1. 2011-06Chromium 12 documents user-specified HSTS preloads.
  2. 2012-11RFC 6797 standardizes HTTP Strict Transport Security.

Sources and what they establish

2012Software
Changed default

With OS X Mountain Lion, Gatekeeper allowed apps from the Mac App Store and from developers identified by Apple by default. A signature let the system check who signed an app and whether it had changed since. Users could tighten or loosen the rule.

The change, in detail
Before
Mac users could download and run applications from many sources without an operating-system default that filtered by publisher identity.
What put it under pressure
Internet-distributed malware made it harder for users to judge where software came from and whether a download had been altered.
What changed
Mountain Lion added Gatekeeper, defaulting to Mac App Store and identified developers, with user-selectable stricter or looser settings.
What followed
Mac software distribution gained a built-in publisher and integrity check. Users could override it, and a valid signature did not establish that software was safe.
Where this applies
OS X Mountain Lion and compatible Macs; settings could be changed and older software was affected by the new default.
Origins and adoption
Apple, introduced during the OS X Mountain Lion release cycle. Shipped in Mountain Lion in July 2012; source documents the default but does not measure how often users overrode it or prevented malware.

Dates in the record

  1. 2012-02-16Apple previewed Gatekeeper and its three selectable policies.
  2. 2012-07-25OS X Mountain Lion became available.

Sources and what they establish

2009–2013Networks
Added defense

Shodan made public services searchable in 2009. In 2013, ZMap showed that one machine on a fast connection could scan public IPv4 addresses for a single service in under 45 minutes. Researchers and security companies used that visibility to measure vulnerable systems and monitor exposure. Attackers could use the same tools to find targets.

The change, in detail
Before
Finding exposed services at Internet scale was slower and more laborious.
What put it under pressure
Unprotected services could remain reachable without their owners noticing.
What changed
Searchable service indexes and high-speed scanning made exposure easier to audit.
What followed
Internet-wide measurement became practical for researchers and security products.
Where this applies
Dual-use visibility tools, not a claim that scanners created vulnerabilities or that all IPv6 space was scanned.
Origins and adoption
Shodan’s service indexing and ZMap’s 2013 research are separate developments. The ZMap authors’ 2024 retrospective documents hundreds of research papers and commercial security adoption.

Dates in the record

  1. 2009Shodan launches its searchable index of public devices and services.
  2. 2013-08ZMap research demonstrated Internet-wide scanning.
  3. 2024Authors documented a decade of research and commercial adoption.

Sources and what they establish

2014–2015Networks
Retired or blocked

An attacker could force some browsers to fall back from TLS to SSL 3.0. The October 2014 POODLE disclosure showed how that downgrade could expose plaintext. Firefox 34 disabled SSL 3.0 that December, and the IETF prohibited the protocol in June 2015.

The change, in detail
Before
Some clients and servers still allowed SSLv3 negotiation or fallback despite TLS having replaced it years earlier.
What put it under pressure
SSLv3's weaknesses accumulated over time, and POODLE showed how downgrade behavior could expose plaintext under practical conditions.
What changed
Firefox disabled SSLv3 in December 2014; the IETF prohibition followed in 2015.
What followed
SSLv3 became a protocol implementations were directed to reject. Older appliances and servers required upgrades or configuration changes.
Where this applies
TLS clients and servers following the standard; legacy endpoints and compatibility products could remain exposed until changed.
Origins and adoption
IETF TLS Working Group, following cryptographic analysis and POODLE disclosure. An enforced Firefox default followed by protocol standardization, not simultaneous removal from all servers.

Dates in the record

  1. 2014-10-14Google disclosed the POODLE attack against SSLv3.
  2. 2015-06RFC 7568 prohibited SSLv3 negotiation and fallback.
  3. 2014-12-01Firefox 34 disabled SSLv3.

Sources and what they establish

2015Devices
Changed default

Android 6.0 let apps ask for sensitive permissions when users tried to use a feature, such as finding their location. Apps targeting the new API had to request those permissions at runtime and handle refusals. Users could also revoke individual permissions later in settings.

The change, in detail
Before
Earlier Android releases grouped permissions at install, so installing or updating an app could grant broad access without a later in-context decision.
What put it under pressure
Users had limited visibility into why an app needed sensitive data and little control over individual grants after installation.
What changed
Android 6.0 introduced runtime requests and individual grant/revoke controls for apps targeting API 23 or higher.
What followed
Permission decisions became visible at point of use and developers had to account for refusal. Prompt timing, habituation, and delayed OS upgrades limited the change's uniform effect.
Where this applies
Android Marshmallow and later; runtime model applies to target API 23+ apps, with legacy target behavior differing.
Origins and adoption
Google's Android platform, introduced with Android 6.0 Marshmallow. Available with Android 6.0 from October 2015, but rollout to existing handsets depended on manufacturers and carriers. No installed-base adoption measure included.

Dates in the record

  1. 2015-05-28Google announced Android M and runtime permissions at Google I/O.
  2. 2015-10-05Android 6.0 Marshmallow began release.

Sources and what they establish

2015–2020Web
Added defense

Getting an HTTPS certificate could mean paying for it and handling renewals by hand. Let’s Encrypt opened its free, automated service to everyone in December 2015, making it easier for small sites to encrypt ordinary pages as well as logins and checkouts. By February 2020 the service reported serving nearly 192 million websites.

The change, in detail
Before
Buy and manually manage a web certificate.
What put it under pressure
Unencrypted browsing exposed traffic to interception.
What changed
Provide free certificates with automated issuance and renewal.
What followed
HTTPS became easier to maintain on small sites.
Where this applies
One enabling change in a broader migration; certificate counts are not unique users.
Origins and adoption
HTTPS predates this service by decades. Let’s Encrypt reported nearly 192 million websites in February 2020.

Dates in the record

  1. 2015-12-03Public beta opened
  2. 2020-02-27Billionth certificate issued

Sources and what they establish

2015–2017Web
Retired or blocked

Java applets and Silverlight needed browser plug-ins, each with its own bugs and updates. As browsers gained built-in alternatives, Chrome dropped the old NPAPI plug-in interface in 2015 and Firefox followed in 2017. Firefox kept a temporary exception for Flash, and enterprise arrangements gave some older applications longer to migrate.

The change, in detail
Before
Websites could load native-code plugins inside the browser, expanding the attack surface and exposing users to separate plugin update cycles.
What put it under pressure
Plugin vulnerabilities, crashes, and maintenance burdens accumulated as browsers added standard APIs for video, audio, and interactive content.
What changed
Browser vendors phased out NPAPI; Firefox 52 removed support for plugins other than Flash in March 2017.
What followed
Java and similar plugins stopped running in mainstream Firefox, and sites moved toward Web APIs. Some enterprise users stayed on older or extended-support versions.
Where this applies
Major desktop browsers and NPAPI plugins; Flash was excluded from the Firefox 52 removal and retired separately later.
Origins and adoption
Browser vendors, with Mozilla announcing its phase-out while web APIs replaced plugin use cases. Removal reached mainstream Firefox in 2017 and Chrome earlier; remaining usage was concentrated in legacy workflows and managed exceptions.

Dates in the record

  1. 2015-10-08Mozilla announced the NPAPI phase-out and schedule.
  2. 2017-03-07Firefox 52 removed NPAPI support other than Flash.

Sources and what they establish

2015–2016Networks
Retired or blocked

RC4 stayed available for compatibility even as attacks against the cipher improved. The IETF prohibited RC4 cipher suites for TLS in 2015, and Firefox 44 disabled them by default in January 2016. Sites that offered only RC4 had to change their encryption settings.

The change, in detail
Before
TLS implementations continued to use RC4 in some cipher suites despite known statistical weaknesses in the stream cipher.
What put it under pressure
Research made plaintext recovery increasingly practical when enough encrypted traffic was available, eroding confidence in RC4 for TLS.
What changed
RFC 7465 prohibited clients from offering and servers from selecting RC4 cipher suites.
What followed
RC4 ceased to be an acceptable TLS choice in standards-compliant implementations, while unpatched and legacy endpoints could persist.
Where this applies
TLS clients and servers; this entry concerns cipher suites, not the retirement of SSL/TLS protocol versions or every other use of RC4.
Origins and adoption
IETF TLS Working Group, with the RFC authored by Alexey Popov at Microsoft. Standards prohibition published in 2015; vendor updates followed, but compliance across all servers was not immediate.

Dates in the record

  1. 2013-03-12Researchers published a practical TLS attack exploiting RC4 biases.
  2. 2015-02RFC 7465 prohibited RC4 cipher suite negotiation.
  3. 2016-01-26Firefox 44 disabled RC4 by default.

Sources and what they establish

2015–2017Web
Retired or blocked

Browsers were already moving away from certificates signed with SHA-1 when researchers demonstrated a practical collision in February 2017. Firefox completed its phase-out for publicly trusted certificate chains that month. Sites still using those certificates needed replacements, while manually imported roots remained an exception.

The change, in detail
Before
Accept SHA-1 signatures in public website certificates.
What put it under pressure
Collision attacks undermined signature security.
What changed
Reject affected certificates in browsers.
What followed
Operators had to migrate to stronger certificate signatures.
Where this applies
Firefox enforcement covered publicly trusted certificate chains. Manually imported roots were exempt.
Origins and adoption
NIST deprecated SHA-1 for new signatures in 2011. Mozilla reported affected sites were under 0.1% of web traffic by February 2017.

Dates in the record

  1. 2015Mozilla announced deprecation plan
  2. 2017-02-23Practical collision announced
  3. 2017-02-24Firefox phase-out reached all users

Sources and what they establish

2017–2021Networks
Changed default

Windows kept its old file-sharing protocol around for compatibility after newer versions were available. WannaCry and other wormable attacks made the cost of leaving SMBv1 available harder to ignore. Starting with Windows 10 and Server version 1709, Microsoft left SMBv1 out of some clean installations by default while keeping SMBv2 and SMBv3 available. Editions and upgrade paths differed, and administrators could still reinstall SMBv1 where supported.

The change, in detail
Before
SMBv1 remained available in Windows installations for legacy file sharing, alongside newer SMB versions.
What put it under pressure
Wormable attacks and exposed file-sharing services made the old protocol a costly compatibility default.
What changed
Microsoft removed SMBv1 from clean installs in stages, while retaining newer SMB versions and some compatibility paths.
What followed
New Windows 11 installs omit client and server components; some earlier editions, upgrades and managed environments retain exceptions.
Where this applies
Microsoft Windows client and server editions; clean installs and upgrades differ.
Origins and adoption
Windows 10 and Windows Server version 1709 began the default transition in 2017. Broad default change on recent clean installs, with legacy and enterprise upgrade exceptions; not universal retirement.

Dates in the record

  1. 2017-10Windows 10 version 1709 and Windows Server version 1709 stop installing SMBv1 by default in several editions.
  2. 2018-10Windows 10 version 1809 removes SMBv1 client by default from clean Pro installs.
  3. 2021-10Windows 11 clean installs omit SMBv1 client and server by default.

Sources and what they establish

2013–2018Web
Added defense

A certificate authority could issue a certificate for your site without giving you an easy way to find out. Certificate Transparency added public logs where domain owners and researchers could look for unexpected certificates. Chrome made logging a condition of trust for publicly trusted certificates issued after April 2018.

The change, in detail
Before
A misissued certificate could be used without a public mechanism for owners or researchers to learn that a certificate authority had issued it.
What put it under pressure
Trust failures and rogue certificate incidents exposed how difficult it was to monitor the global certificate authority system centrally.
What changed
The IETF specified public, append-only Certificate Transparency logs, with monitors and browser policies encouraging or requiring logged certificates.
What followed
Issuance gained a public audit trail and monitoring ecosystem; detection still depends on log coverage and active monitoring.
Where this applies
Public Web PKI certificates and participating logs, monitors, and browsers; it does not cover all private PKI or block issuance itself.
Origins and adoption
The first CT RFC was published in 2013; browser enforcement expanded later. Chrome requires CT for publicly trusted certificates issued after April 30, 2018.

Dates in the record

  1. 2013-06RFC 6962 specified CT.
  2. 2018-04-30Certificate issuance cutoff for Chrome’s general CT requirement.

Sources and what they establish

2018–19Web
Added defense

Chrome used to let pages from different sites share a renderer process. Spectre raised the urgency of work already underway to separate them. In July 2018, Google reported that Chrome 67 had enabled Site Isolation for 99% of supported desktop users. Chrome 77 later extended a narrower version to sites where users signed in on eligible Android devices.

The change, in detail
Before
Different websites could share a browser renderer process while relying on software rules to keep their data separate.
What put it under pressure
Spectre and renderer bugs demonstrated that process sharing could expose cross-site information.
What changed
Chrome 67 separated cross-site documents into different renderer processes for most supported desktop users. Chrome 77 expanded the protection against compromised renderers on desktop and introduced narrower Android coverage.
What followed
Desktop site isolation became a default; Android coverage is limited by device memory and site sign-in conditions.
Where this applies
Chrome desktop platforms from version 67; Android devices with at least 2GB RAM from version 77 for signed-in sites.
Origins and adoption
Desktop default shipped in May 2018; Android expansion followed in 2019. Google reported 99% of supported desktop users in July 2018. Chrome 77 covered sites where users signed in on eligible Android devices.

Dates in the record

  1. 2018-01Spectre disclosure increases urgency of the existing Site Isolation project.
  2. 2018-07-11Google reports Site Isolation enabled for 99% of supported Chrome 67 desktop users.
  3. 2019-09Chrome 77 adds further desktop protections and isolates sites where users sign in on eligible Android devices.

Sources and what they establish

2018Web
Changed default

In July 2018, Chrome 68 began labeling HTTP pages “Not secure” in the address bar. Anyone who could observe or intercept an HTTP connection could read or change the traffic. Google had already counted 81 of the top 100 sites using HTTPS by default earlier that year.

The change, in detail
Before
Browse ordinary pages over unencrypted HTTP.
What put it under pressure
Network observers could read or alter the traffic.
What changed
Label every HTTP page “Not secure.”
What followed
Lack of encryption became visible in everyday browsing.
Where this applies
Chrome 68 interface change; HTTP remained accessible.
Origins and adoption
Earlier Chrome versions warned on selected sensitive HTTP pages. Google reported 81 of the top 100 sites used HTTPS by default in February 2018.

Dates in the record

  1. 2018-02-08Chrome announced broader warning
  2. 2018-07-24Chrome 68 stable release

Sources and what they establish

2019–2023Networks
Changed default

A flaw that lets attackers make a server fetch an arbitrary URL can expose the temporary credentials available from an EC2 instance’s metadata service. AWS introduced IMDSv2 in 2019, requiring a session token for metadata requests. In November 2023, AWS made IMDSv2 the default for new launches through the console’s Quick Start options.

The change, in detail
Before
EC2 software could fetch metadata and temporary role credentials through a simple link-local request.
What put it under pressure
SSRF flaws and misconfigured proxies can let an attacker reach internal metadata endpoints.
What changed
AWS introduced token-based IMDSv2 and made it the default for new Quick Start launches.
What followed
Console Quick Start launches gained an IMDSv2-only default. Existing instances can still require a customer migration.
Where this applies
Amazon EC2 metadata service; default varies by AMI, instance type, launch pathway and account policy.
Origins and adoption
IMDSv2 introduced November 2019; default migration milestones continued through mid-2024. AWS confirmed the default for console Quick Start launches in November 2023.

Dates in the record

  1. 2019-11-19AWS introduces IMDSv2.
  2. 2023-11AWS switches console Quick Start launches to IMDSv2-only.

Sources and what they establish

2019Identity
Added defense

Attackers try username and password pairs from one breach on other sites. Google’s Password Checkup began warning users about exposed credentials through a Chrome extension in 2019. In a study of that extension, users changed their passwords after 26% of warnings. Google later added the check to Chrome.

The change, in detail
Before
Reuse a password without knowing it had leaked elsewhere.
What put it under pressure
Attackers tried stolen credentials across other services.
What changed
Compare logins against known compromised credentials.
What followed
Warnings prompted some users to replace exposed passwords.
Where this applies
The study measured extension users; warnings did not automatically fix accounts.
Origins and adoption
Password breach services existed before browser integration. The 2019 study observed 21 million logins from nearly 670,000 users.

Dates in the record

  1. 2019-02Password Checkup extension launched
  2. 2019-12-10Chrome integration announced

Sources and what they establish

Back to filters
2020–2026

Changes you didn’t have to turn on

A phone could take back access from an app you no longer used, and signing in could mean using the same screen lock that opened your device. Package registries tightened publishing credentials, and new cloud storage buckets became harder to expose by accident.

2020Devices
Changed default

Android 11 began resetting sensitive permissions for apps targeting Android 11 or later after several months of disuse. It also introduced one-time grants for location, camera, and microphone access. People could give an app access for one use and grant it again when needed.

The change, in detail
Before
Runtime permissions granted to an app could remain in place after a user stopped opening it.
What put it under pressure
Dormant apps retain access to personal data and device features after their active use has ended.
What changed
Android 11 introduced automatic permission reset for long-unused apps and one-time grants for sensitive permissions.
What followed
Permissions can be revoked automatically after months of inactivity; exclusions and user controls remain.
Where this applies
Android 11+ runtime permissions; exact availability varies by app eligibility and device implementation.
Origins and adoption
Android 11 shipped auto-reset and one-time grants in 2020. At launch, automatic reset targeted unused apps built for Android 11 or later.

Dates in the record

  1. 2020-09Android 11 ships one-time permission grants and auto-reset for eligible unused apps.

Sources and what they establish

2020Web
Retired or blocked

Older versions of TLS lacked modern cryptographic options. Browser makers gave site operators advance notice that they were moving to TLS 1.2 or later. In June 2020, Firefox 78 disabled TLS 1.0 and 1.1 by default, and sites relying on them began showing an error page.

The change, in detail
Before
Browsers continued to negotiate TLS 1.0 and 1.1 for compatibility with older servers.
What put it under pressure
Older versions lacked modern cryptographic options and retained known design weaknesses; most connections already used TLS 1.2 or later.
What changed
Firefox 78 disabled TLS 1.0 and 1.1 by default after coordinated browser deprecation announcements.
What followed
Firefox users encountered errors on sites unable to use TLS 1.2 or later.
Where this applies
Firefox HTTPS connections and related browser deprecation announcements.
Origins and adoption
Mozilla announced the plan in October 2018 and confirmed disablement in the Firefox 78 release notes in June 2020. The Firefox 78 release notes confirm shipped disablement. Browser vendors used different schedules.

Dates in the record

  1. 2018-10-15Mozilla announces its TLS 1.0 and 1.1 deprecation plan.
  2. 2020-06-30Firefox 78 disables TLS 1.0 and TLS 1.1 by default.

Sources and what they establish

2017–2021Web
Retired or blocked

In 2014, 80% of desktop Chrome users visited a site containing Flash each day. Google put that figure at 17% in 2017 as sites moved to built-in web technologies. Adobe ended support for Flash Player on December 31, 2020, and blocked Flash content from running in the player on January 12, 2021.

The change, in detail
Before
People installed a browser plug-in to play web video, games and applications.
What put it under pressure
Security exposure persisted while native web alternatives matured.
What changed
Browser vendors removed Flash support, and Adobe blocked content from running in its general-distribution player.
What followed
Flash pages no longer worked in ordinary supported browsers.
Where this applies
Adobe’s general-distribution player; specialist legacy arrangements and emulators differ.
Origins and adoption
Flash was a major web platform for roughly two decades. Google reported that the share of desktop Chrome users visiting a site containing Flash each day fell from 80% in 2014 to 17% in 2017.

Dates in the record

  1. 2017-07Retirement announced
  2. 2020-12-31Adobe support ended
  3. 2021-01-12Adobe blocked content execution

Sources and what they establish

2021–2022Identity
Changed default

Google began turning on two-step verification automatically for eligible accounts in 2021. It reported enrolling more than 150 million people and seeing a 50% drop in compromised accounts after the rollout.

The change, in detail
Before
Protect an account with a password unless the user opted into more.
What put it under pressure
Breached passwords and phishing enabled account takeovers.
What changed
Automatically enroll eligible accounts in two-step verification.
What followed
Many more users encountered a second check at sign-in.
Where this applies
Google’s eligible accounts and vendor-reported outcome; MFA methods differ.
Origins and adoption
Two-step verification long predates automatic enrollment. Google reported over 150 million auto-enrollments in 2021.

Dates in the record

  1. 2021Mass automatic enrollment
  2. 2022-02-08Google reported results

Sources and what they establish

2022–2023Identity
Retired or blocked

Older Exchange clients could use a reusable account password for each request. Microsoft began turning off Basic Authentication for protocols including POP, IMAP, and Exchange ActiveSync in October 2022. It disabled the protocols for remaining extension tenants starting in early 2023, sending affected clients to OAuth. SMTP AUTH followed a separate schedule.

The change, in detail
Before
Apps and clients could authenticate to Exchange Online with a username and password on each request.
What put it under pressure
Stored and repeatedly transmitted credentials increased exposure to password theft and account takeover.
What changed
Microsoft shifted supported clients to OAuth-based Modern Authentication and disabled Basic Auth protocol by protocol.
What followed
Basic Auth cannot be re-enabled for affected protocols. Affected integrations had to migrate.
Where this applies
Affected Exchange Online protocols and tenants. SMTP AUTH and on-premises Exchange follow separate policies.
Origins and adoption
Microsoft announced deprecation in 2019 and completed staged disablement starting October 2022. Microsoft confirmed enforcement across tenants for the protocols covered by this rollout.

Dates in the record

  1. 2019-09-20Microsoft announces Exchange Online Basic Auth deprecation.
  2. 2022-10-01Permanent tenant disablement rollout begins.
  3. 2022-12-31Tenant re-enable window ends.
  4. 2023Microsoft disables Basic Authentication for remaining extension tenants on affected protocols.

Sources and what they establish

2022–2023Software
Changed default

Malicious Office attachments often asked people to click “Enable Content” to run a macro. Starting in 2022, Microsoft rolled out a default block on VBA macros in files Windows marked as coming from the internet. The block reached Office update channels into 2023. Organizations could still use trusted files and locations.

The change, in detail
Before
Click “Enable Content” to run a downloaded document’s macros.
What put it under pressure
Phishing documents persuaded users to enable malicious VBA.
What changed
Block internet-origin macros by default.
What followed
The familiar enable button no longer bypassed this block.
Where this applies
Affected Office apps on Windows, with origin marks and trust exceptions; not a ban on all macros.
Origins and adoption
Malicious macros date to the 1990s. NCSC describes macros as widely embedded in business workflows and attacks.

Dates in the record

  1. 2022-07-27Current Channel rollout started
  2. 2023-01-10Semi-Annual Enterprise Channel change

Sources and what they establish

2022–2024Software
Still changing

Memory errors in C and C++ can give attackers a way to compromise a device. In 2022, Google reported that Android 13 shipped with a majority of newly added code written in memory-safe languages, and that Rust accounted for about 21% of new native code. Android used Rust in components such as Keystore2 and its ultra-wideband stack. Google continued to harden its existing C++ code.

The change, in detail
Before
Android used C and C++ for many low-level components.
What put it under pressure
Memory safety bugs remained a significant source of exploitable Android vulnerabilities.
What changed
Google used Rust for selected new Android components and hardened existing C++.
What followed
Android 13 shipped Rust in Keystore2, ultra-wideband, and other components.
Where this applies
Android platform code and Google engineering practice.
Origins and adoption
Google documented Android 13 adoption in 2022 and subsequent expansion and hardening work in 2024. Google reported that a majority of new Android 13 code used memory-safe languages, with Rust at about 21% of new native code.

Dates in the record

  1. 2022-12Google documents memory-safe code and Rust components shipped in Android 13.
  2. 2024-10Google describes further adoption and continued C++ hardening.

Sources and what they establish

2023–24Identity
Added defense

A stolen developer account can give attackers a way to publish code under a maintainer’s name. GitHub began requiring two-factor authentication for groups of code contributors in March 2023. It reported that nearly 95% of contributors notified that year enrolled, and adoption among all active contributors rose 54% relative to its previous level.

The change, in detail
Before
Code contributors could maintain GitHub accounts with password-only authentication.
What put it under pressure
Social engineering and account takeover create routes to malicious code changes and releases.
What changed
GitHub progressively required contributors and administrators in selected groups to enroll in 2FA.
What followed
GitHub reported nearly 95% enrollment among contributors notified in 2023 and 54% higher adoption across active contributors.
Where this applies
GitHub.com contributors and selected platform populations; staged program and vendor-reported measurement.
Origins and adoption
Rollout began March 13, 2023, after a 2022 commitment to require 2FA by the end of 2023. Strong within notified GitHub cohorts; not every user or software repository globally.

Dates in the record

  1. 2023-03-13GitHub starts staged enrollment requirements.
  2. 2023-12Initial rollout target for contributors is reached.
  3. 2024-03GitHub reports nearly 95% notified-cohort enrollment and 54% rise among active contributors.

Sources and what they establish

2023–24Identity
Added defense

Passkeys use a credential tied to a specific service and unlocked with a device PIN or biometric. In October 2023, Google made passkeys the default sign-in option for personal Google Accounts on supported devices, while keeping passwords available. By May 2024, Google reported more than a billion authentications across more than 400 million accounts.

The change, in detail
Before
Online accounts generally presented passwords as the main sign-in method, with other factors often added afterward.
What put it under pressure
Phishing and reused or stolen passwords remain common account compromise paths.
What changed
FIDO/WebAuthn passkeys use public-key credentials unlocked by a device PIN or biometric; Google changed the prompt order on eligible devices.
What followed
Google reported 1B+ passkey authentications across 400M+ accounts within a year; password fallback persists.
Where this applies
Google Accounts on supported devices and services implementing passkeys; this is not web-wide password retirement.
Origins and adoption
Google announced default prompts on October 10, 2023 after enabling account support in May. High measured use in Google's accounts, but no comparable ecosystem-wide adoption measure in these sources.

Dates in the record

  1. 2023-05Google Accounts begin supporting passkey sign-in.
  2. 2023-10-10Passkeys become the default suggested sign-in method on supported devices.
  3. 2024-05-02Google reports more than one billion authentications across 400 million accounts.

Sources and what they establish

2018–2023Software
Changed default

S3 buckets started private, but a mistaken access setting could still expose the files inside. AWS added Block Public Access in 2018, then switched it on and disabled access control lists for newly created buckets in April 2023. Making a new bucket public now takes an extra step; the rollout left existing buckets as they were.

The change, in detail
Before
A policy or ACL could make a private bucket publicly accessible.
What put it under pressure
Mistakes in access settings exposed stored data.
What changed
Enable Block Public Access and disable ACLs on new buckets.
What followed
Public sharing required an explicit configuration change.
Where this applies
New buckets across AWS Regions; existing buckets were not retroactively changed.
Origins and adoption
S3 always started private; the additional guard arrived in 2018. Default coverage applies to every newly created bucket, not a measured share of existing storage.

Dates in the record

  1. 2018-11-15Block Public Access launched
  2. 2023-04-05New default rollout began
  3. 2023-04-27AWS reported completion in all Regions

Sources and what they establish

2024–25Networks
Changed default

Gmail began requiring senders to authenticate mail sent to personal accounts in 2024. Bulk senders faced additional authentication requirements and had to offer one-click unsubscribe for marketing and subscribed messages. Google began increasing temporary and permanent rejections of mail that failed its rules in November 2025.

The change, in detail
Before
Email delivery across providers accepted inconsistent sender authentication and bulk-sender practices.
What put it under pressure
Forged sender identities and spam campaigns eroded recipient trust and enabled phishing.
What changed
Google tied delivery to baseline sender authentication and stricter bulk-sender requirements, then ramped rejection enforcement.
What followed
Requirements apply to mail reaching personal Gmail; Google documents temporary and permanent rejections beginning November 2025.
Where this applies
Personal Gmail accounts; senders over 5,000 messages per day face additional requirements. Other mailbox providers set their own policies.
Origins and adoption
Requirements announced October 2023 and took effect February 1, 2024; stronger rejection enforcement began November 2025. Large receiving-platform policy with broad sender impact, but no overall compliance rate published in reviewed sources.

Dates in the record

  1. 2023-10-03Google announces new bulk sender requirements.
  2. 2024-02-01Sender requirements take effect for personal Gmail delivery.
  3. 2025-11Google ramps up temporary and permanent rejection of non-compliant traffic.

Sources and what they establish

2024 onwardSoftware
Changed default

Attackers could use a document to persuade people to enable embedded ActiveX controls. Office LTSC 2024 disabled the controls by default, and Microsoft now documents the same default for Microsoft 365 and Office 2024 in Word, Excel, PowerPoint, and Visio. Users can change the setting in Trust Center.

The change, in detail
Before
Windows desktop Office supported ActiveX controls and some configurations allowed them to run in documents.
What put it under pressure
Malicious attachments and deceptive prompts can trick users into enabling risky embedded controls.
What changed
Microsoft changed the default setting to disable ActiveX controls in Microsoft 365 and Office 2024.
What followed
Controls cannot be created or interacted with under the default setting; administrators or users can still enable them.
Where this applies
Microsoft 365 and Office 2024 desktop applications on Windows; Mac and web Office do not support ActiveX.
Origins and adoption
Office LTSC 2024 shipped with the controls disabled by default. Current Microsoft support confirms the same default for Microsoft 365 and Office 2024. Default across current supported desktop Office versions, with explicit override and legacy behavior remaining.

Dates in the record

  1. 2024Office LTSC 2024 disables ActiveX controls by default.

Sources and what they establish

2024Software
Changed default

A stolen PyPI account can let someone publish a package under a name developers already trust. PyPI began requiring two-factor authentication for everyone who manages projects or uploads releases. You can still browse and install packages without signing in, and automated publishing can use API tokens or trusted publishing.

The change, in detail
Before
PyPI accounts could perform project management and release uploads without universal 2FA enrollment.
What put it under pressure
Compromised maintainer accounts can be used to upload malicious package versions.
What changed
PyPI required all account holders to enable 2FA before management actions and uploads, following earlier staged requirements.
What followed
Mandatory account 2FA went into effect January 1, 2024; browsing and package downloads remain available without accounts.
Where this applies
All PyPI accounts and projects; package consumers without accounts are unaffected.
Origins and adoption
PyPI added 2FA in 2019 and began enforcing it in stages before January 2024. Registry-wide rule, but no adoption metric is supplied; this does not cover all language registries.

Dates in the record

  1. 2019PyPI adds support for two-factor authentication.
  2. 2023-12-13PyPI announces universal 2FA requirement beginning January 1.
  3. 2024-01-012FA gate applies to management actions and uploads.

Sources and what they establish

2025–2026Software
Retired or blocked

A token stolen from a developer’s computer or a build system could let someone keep publishing packages under that developer’s name. In December 2025, npm revoked its classic tokens and limited how long newer write-enabled tokens could last. In July 2026, tokens that bypassed two-factor authentication lost the ability to manage accounts, organizations and packages without an interactive check. Interactive logins last two hours, and build systems still have exceptions for granular tokens.

The change, in detail
Before
npm publishers could rely on classic long-lived tokens with broad access and no short expiry.
What put it under pressure
Compromised package-maintainer accounts and stolen registry credentials have enabled malicious package releases.
What changed
npm revoked classic tokens and shortened interactive login sessions. In July 2026, sensitive management actions began requiring interactive 2FA even when a token was configured to bypass it.
What followed
Classic tokens no longer authenticate; existing publisher workflows had to migrate. In July 2026, governance actions using bypass-2FA tokens began requiring interactive 2FA. Granular tokens and trusted publishing remain available.
Where this applies
npm registry credentials and package maintainers; does not affect GitHub tokens or all package registries.
Origins and adoption
Security changes announced September 29, 2025, followed by staged rollout and final classic-token revocation December 9. Enforced across npm classic tokens; token rotation burden and automation exceptions remain. This is a registry-specific outcome.

Dates in the record

  1. 2025-09-29GitHub announces phased npm token lifetime and classic-token changes.
  2. 2025-11-05Classic token creation is disabled and write-token defaults/lifetimes change.
  3. 2025-12-09npm permanently revokes classic tokens and changes interactive login to two-hour sessions.
  4. 2026-07-31Tokens configured to bypass 2FA lose sensitive account, organization and package management actions unless the user completes interactive 2FA.

Sources and what they establish

2026Web
Changed default

Chrome used to show “Not secure” after it had already connected over HTTP. In October 2026, Google reported that Chrome now asks before connecting to a public site if HTTPS fails. The browser remembers an approval until the site goes unvisited for two weeks. Local addresses and intranets are generally exempt.

The change, in detail
Before
Connect over HTTP, then show a warning in the address bar.
What put it under pressure
An attacker could intercept the first request, including a redirect to HTTPS.
What changed
Ask before making an insecure connection to a public site.
What followed
Public HTTP access became a deliberate exception.
Where this applies
Chrome default; bypasses, remembered approvals and local-network exceptions remain.
Origins and adoption
An opt-in secure-connection setting preceded the default. Google’s rollout report says nearly 95% of desktop users saw no HTTP warning in a given week.

Dates in the record

  1. 2025-10-28Default change announced
  2. 2026-10-21Google reported default rollout completed

Sources and what they establish

Back to filters

About this history

This collection follows changes in everyday and administrative practice: what people trusted, what exposed the risk, and what was actually retired, blocked or deployed. An attack alone wasn’t enough to make the list.

Some changes came from a single urgent incident. Others took years of research, product decisions and changing expectations. “Added defense” records a deployed response; “Still changing” marks a transition with a substantial part of the older practice remaining. Neither means the threat disappeared.

The entries are a selection, with an emphasis on PCs, widely used browsers and large internet services. They aren’t a complete global history. Sources include original papers, standards, release notes and vendor reports; adoption claims are limited to the populations those sources describe.

Researched through November 6, 2026. Research used AI assistance, including separate factual and writing reviews. The source notes and review record are public so the claims can be checked and corrected.

Read the research · Suggest a correction or an addition