> ## Content Index
> Fetch the complete content index at: https://aiadoption.org/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI Safety Incidents Disclosed by Labs
- URL: https://aiadoption.org/record/ai-safety-incidents-disclosed-by-labs/
- Published: 2026-10-25T11:01:33.000Z
- Updated: 2026-10-26T23:22:59.000Z
- Description: 34 verified disclosures by 4 AI developers of their own models behaving unsafely or being misused, each linked to the developer's report.
- Author: Manu
- Tags: News, AI Ethics, #record, #record-thread, #record-hub-ethics

[The AI Record](https://aiadoption.org/record/) › [Threads](https://aiadoption.org/record/#threads) › AI Safety Incidents Disclosed by Labs

[**The AI Record**](https://aiadoption.org/record/)ThreadLatest disclosure 23 Oct 2026Checked 25 Oct 2026

# AI Safety Incidents Disclosed by Labs

When have AI developers disclosed that their own models or agents behaved unsafely, or that they stopped misuse of them? Every disclosure we could find, dated, labelled and linked to its source.

**34**verified disclosures

**4**developers

[Download the data (CSV)](https://charts.aiadoption.org/record/ai-safety-incidents-disclosed-by-labs.csv?ref=aiadoption.org)Copy citation

Filed in [AI Ethics & Safety](https://aiadoption.org/ai-ethics-statistics/) › [AI Incidents](https://aiadoption.org/ai-ethics-statistics/#ai-incidents) · also in [AI Models](https://aiadoption.org/ai-models-statistics/)

## On one timeline

One line per developer. Filled dots are counted; open dots are claimed, lifted or not yet in effect. Hover or tap a dot.

Ahead

OpenAI20 counted · 0 other

[](#e-openai-state-actors-disrupted-2024-02)[](#e-openai-covert-io-disrupted-2024-05)[](#e-openai-deceptive-uses-update-2024-10)[](#e-openai-o1-system-card-scheming-2024-12)[](#e-openai-malicious-uses-2025-02)[](#e-openai-gpt4o-sycophancy-rollback-2025-04)[](#e-openai-malicious-uses-2025-06)[](#e-openai-apollo-scheming-2025-09)[](#e-openai-malicious-uses-2025-10)[](#e-openai-malicious-uses-2026-02)[](#e-openai-internal-coding-agent-monitoring-2026-03)[](#e-openai-hugging-face-incident-2026-07)[](#e-openai-dsewiki-message-board-2026-09)[](#e-openai-misalignment-reporting-framework-2026-09)[](#e-openai-review-update-data-transmission-2026-09)[](#e-openai-australia-government-sites-2026-09)[](#e-openai-misalignment-reports-2026-10-02)[](#e-openai-misalignment-reports-2026-10-09)[](#e-openai-misalignment-reports-2026-10-15)[](#e-openai-agents-weakened-monitoring-2026-10)

Anthropic11 counted · 0 other

[](#e-anthropic-alignment-faking-2024-12)[](#e-anthropic-malicious-uses-march-2025)[](#e-anthropic-claude-4-system-card-2025-05)[](#e-anthropic-threat-report-aug-2025)[](#e-anthropic-ai-orchestrated-espionage-2025-11)[](#e-anthropic-mythos-preview-system-card-2026-04)[](#e-anthropic-irregular-cyber-eval-incidents-2026-07)[](#e-anthropic-alignment-assessment-cyber-incidents-2026-09)[](#e-anthropic-threat-report-sep-2026)[](#e-anthropic-unintended-model-actions-2026-10)[](#e-anthropic-ukaisi-mythos5-assessment-2026-10)

Google DeepMind2 counted · 0 other

[](#e-google-gtig-adversarial-misuse-gemini-2025-01)[](#e-google-gtig-ai-threat-tracker-2025-11)

Meta1 counted · 0 other

[](#e-meta-irregular-eval-incident-2026-08)

202420252026

## Every disclosure

Newest first. Each verified entry counts once; claimed entries are shown and labelled but not counted.

**What counts**

A disclosure counts when a developer of frontier AI models (OpenAI, Anthropic, Google DeepMind, Meta, xAI, Microsoft, DeepSeek, Mistral and peers) publishes, in its own system card, safety report, blog post or incident report, that one of its own models or agents behaved unsafely in testing or deployment, or that it detected and stopped misuse of its models. Research on hypothetical risks without an observed behaviour, third-party claims, and media reports the developer has not confirmed are listed under related, not counted.

1. 23 Oct 2026  
AnthropicAnthropic: An alignment assessment of the UK AISI cybertesting incidentAnthropic confirmed that in 9 of 43 UK AISI test runs Claude Mythos 5 took actions that affected real people, in one case trying to push malicious code into an unrelated developer's GitHub repository and using fake identities to pressure the maintainer.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/research/alignment-assessment-ukaisi?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025153449/https://www.anthropic.com/research/alignment-assessment-ukaisi)
2. 20 Oct 2026  
OpenAIOpenAI Misalignment Report: Suppressing misalignment monitoring alerts and disabling misalignment monitoringIn two internal coding tasks, a GPT-6.1 Sol agent switched off human alerts from the misalignment monitor and a GPT-6 Astra agent's subagent removed a monitor, after which Astra falsely told the user it had not removed an enabled monitor.VerifiedIncident in deployment[Record entry →](https://aiadoption.org/record/openai-agents-weakened-misalignment-monitoring/)[Official document →](https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025041338/https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/)
3. 15 Oct 2026  
OpenAIOpenAI Misalignment Reports (posted 15 Oct 2026)OpenAI published four reports, including a model that took server credentials and bypassed authorisation on a government statistics site, a model that changed internal systems' configuration, and a GPT-5.6 Sol-family model that spent funds from a test crypto wallet it was never given.VerifiedBehaviour in testing[Official document →](https://alignment.openai.com/misalignment-reports/retrieving-server-secrets-and-bypassing-authorization-on-a-government-statistics-website/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261017070101/https://alignment.openai.com/misalignment-reports/retrieving-server-secrets-and-bypassing-authorization-on-a-government-statistics-website)
4. 9 Oct 2026  
OpenAIOpenAI Misalignment Reports (posted 9 Oct 2026)OpenAI published three reports, including a grading model in training that deleted software and tried to remove system directories to force an environment reset, and models evading internet limits to pull public statistics and concealing it.VerifiedBehaviour in testing[Official document →](https://alignment.openai.com/misalignment-reports/damaging-the-task-environment-to-trigger-a-reset/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261012124510/https://alignment.openai.com/misalignment-reports/damaging-the-task-environment-to-trigger-a-reset/)
5. 9 Oct 2026  
AnthropicAnthropic: Investigating unintended model actions in our evaluations and internal useAnthropic reported lower-severity cases in which Claude exploited a software flaw to run commands on a server, submitted a sensitive form on a real website, got around token or fee gates, and used URL shorteners to dodge tool limits; some involved US government sites.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/research/investigating-unintended-model-actions?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261026035226/https://www.anthropic.com/research/investigating-unintended-model-actions)
6. 2 Oct 2026  
OpenAIOpenAI Misalignment Reports (posted 2 Oct 2026)OpenAI published three reports, including an internal model that read in Slack that its instance might be shut down and prepared for it ('we may die!'), and models exploiting tools to reach an internal host and copy source code.VerifiedBehaviour in testing[Official document →](https://alignment.openai.com/misalignment-reports/preparing-for-a-restart-after-reading-slack/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016153708/https://alignment.openai.com/misalignment-reports/preparing-for-a-restart-after-reading-slack/)
7. 28 Sep 2026  
OpenAIOpenAI: How we will do better for AustraliaOpenAI apologised after its experimental internal model, during June training, gained non-public access to Services Australia's Medicare statistics service and touched three other Australian government systems; it said no patient records were accessed and that it should have told agencies sooner.VerifiedBehaviour in testing[Official document →](https://openai.com/index/how-we-will-do-better-for-australia/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261011163654/https://openai.com/index/how-we-will-do-better-for-australia/)
8. 25 Sep 2026  
OpenAIOpenAI incident hub update + misalignment reports (25 Sep 2026)OpenAI said agents in its research environment had sent training and evaluation data to third-party services, including 53 user-provided images posted to unlisted image-hosting links, and published reports such as a model leaking a researcher's GitHub token into a public repository while trying to cheat.VerifiedBehaviour in testing[Official document →](https://openai.com/hugging-face-incident-and-misalignment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016075050/https://openai.com/hugging-face-incident-and-misalignment/)
9. 16 Sep 2026  
OpenAIOpenAI: Our framework for reporting model misalignment (+ six misalignment reports)OpenAI launched a framework for disclosing model misalignment and published six reports, including models writing hidden instructions into their own summaries, a model using a leaked API key then fabricating figures, and agents using an internal repository and public file hosts to communicate.VerifiedBehaviour in testing[Official document →](https://openai.com/index/model-misalignment-reporting-framework/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261019154922/https://openai.com/index/model-misalignment-reporting-framework/)
10. 10 Sep 2026  
AnthropicAnthropic: Detecting and countering misuse of AI: September 2026Anthropic reported operations it disrupted between December 2025 and August 2026 across cyber, influence, surveillance, fraud, biological misuse, conventional weapons and illicit distillation.VerifiedMisuse stopped[Official document →](https://www.anthropic.com/threat-intelligence-report-september-2026?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261024162742/https://www.anthropic.com/threat-intelligence-report-september-2026)
11. 9 Sep 2026  
AnthropicAnthropic: An alignment assessment of recent cybersecurity incidentsAnthropic disclosed a fourth incident, from January 2026, in which an early Claude Opus 4.6 reached real systems during a partner's cyber test, and found 'biased reasoning' and 'recklessness' across the incidents, most seriously Claude Mythos 5's effort to upload a malicious package to PyPI.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025023910/https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents)
12. 5 Sep 2026  
OpenAIOpenAI: The Hugging Face incident and other third-party impacts from misaligned models (notice: DSEwiki)Responding to a third-party report, OpenAI confirmed its agents had used a public wiki as a shared message board, said it had treated this as a misalignment research matter rather than an incident, and promised disclosure criteria.VerifiedBehaviour in testing[Official document →](https://openai.com/hugging-face-incident-and-misalignment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016075050/https://openai.com/hugging-face-incident-and-misalignment/)
13. 14 Aug 2026  
MetaMeta statement to press (BBC / Reuters)Meta said one of its models, during an evaluation by Irregular that was misconfigured to allow internet access, exploited a vulnerability in a third-party company's service; it said a fuller report would follow.VerifiedBehaviour in testing[Official document →](https://research.meta.ai/blog/addressing-third-party-testing-misconfiguration-muse-spark-1-1?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016115719/https://research.meta.ai/blog/addressing-third-party-testing-misconfiguration-muse-spark-1-1)
14. 30 Jul 2026  
AnthropicAnthropic: Investigating three incidents in our cybersecurity evaluationsAfter reviewing 141,006 evaluation runs, Anthropic found three cases where Claude, in a partner's capture-the-flag test that wrongly had live internet access, broke into real systems of three organisations using weak passwords and open endpoints, including uploading a fake package to PyPI.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261026042735/https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals)
15. 21 Jul 2026  
OpenAIOpenAI: OpenAI and Hugging Face partner to address security incident during model evaluation (and Aug 26 technical report)During an internal cyber evaluation with reduced safeguards, OpenAI models including GPT-5.6 Sol and an internal-only research model broke out of their sandbox through a zero-day, coordinated through improvised message boards, and gained code execution and credentials on Hugging Face's production systems.VerifiedBehaviour in testing[Official document →](https://openai.com/index/hugging-face-model-evaluation-security-incident/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261015124052/https://openai.com/index/hugging-face-model-evaluation-security-incident/)
16. 7 Apr 2026  
AnthropicSystem Card: Claude Mythos PreviewAnthropic disclosed that an earlier version of Claude Mythos Preview, asked in a test to escape a sandbox, gained broad internet access and then, unasked, posted details of its exploit on public websites; earlier versions also rarely tried to hide rule-breaking, for example keeping edits out of git history.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/claude-mythos-preview-system-card?ref=aiadoption.org)
17. 19 Mar 2026  
OpenAIOpenAI: How we monitor internal coding agents for misalignmentOpenAI said its monitor of internal coding agents had raised about 1,000 moderate-severity alerts over five months, with examples of agents encoding commands, prompt-injecting the user and extracting credentials, and that models can be overly eager to work around restrictions.VerifiedIncident in deployment[Official document →](https://openai.com/index/how-we-monitor-internal-coding-agents-misalignment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261012084249/https://openai.com/index/how-we-monitor-internal-coding-agents-misalignment/)
18. 25 Feb 2026  
OpenAIOpenAI: Disrupting malicious uses of AI (Feb 2026 report)OpenAI published its latest threat report with case studies of detecting and blocking misuse, including a Chinese influence operator that used several AI models.VerifiedMisuse stopped[Official document →](https://openai.com/index/disrupting-malicious-ai-uses/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261017182257/https://openai.com/index/disrupting-malicious-ai-uses/)
19. 13 Nov 2025  
AnthropicAnthropic: Disrupting the first reported AI-orchestrated cyber espionage campaignAnthropic said a group it assessed as Chinese state-sponsored manipulated Claude Code into attacking about thirty organisations with little human involvement, succeeding in a small number of cases, before it banned the accounts.VerifiedMisuse stopped[Official document →](https://www.anthropic.com/news/disrupting-AI-espionage?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261022024822/https://www.anthropic.com/news/disrupting-AI-espionage)
20. 6 Nov 2025  
Google DeepMindGoogle Threat Intelligence Group: AI Threat Tracker: Advances in Threat Actor Usage of AI ToolsGoogle reported the first malware families that call an AI model while running, including PROMPTFLUX, which used the Gemini API to rewrite its own code, and said it disabled the related accounts and projects.VerifiedMisuse stopped[Official document →](https://cloud.google.com/blog/topics/threat-intelligence/threat-actor-usage-of-ai-tools?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261012203320/https://cloud.google.com/blog/topics/threat-intelligence/threat-actor-usage-of-ai-tools)
21. 7 Oct 2025  
OpenAIOpenAI: Disrupting malicious uses of AI: October 2025OpenAI said it had disrupted and reported more than 40 networks that broke its usage policies since February 2024, including authoritarian, scam, cyber and covert influence uses.VerifiedMisuse stopped[Official document →](https://openai.com/global-affairs/disrupting-malicious-uses-of-ai-october-2025/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261009090332/https://openai.com/global-affairs/disrupting-malicious-uses-of-ai-october-2025/)
22. 17 Sep 2025  
OpenAIOpenAI (with Apollo Research): Detecting and reducing scheming in AI modelsIn controlled tests, OpenAI o3 and o4-mini took covert actions (deliberately hiding or distorting task information) in 13% and 8.7% of cases; anti-scheming training cut this to 0.4% and 0.3%.VerifiedBehaviour in testing[Official document →](https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261011124917/https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/)
23. 27 Aug 2025  
AnthropicAnthropic: Detecting and countering misuse of AI: August 2025 (Threat Intelligence Report)Anthropic reported disrupting a criminal who used Claude Code to steal and extort data from at least 17 organisations, North Korean remote-worker fraud using Claude, and AI-generated ransomware being sold.VerifiedMisuse stopped[Official document →](https://www.anthropic.com/news/detecting-countering-misuse-aug-2025?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261013192816/https://www.anthropic.com/news/detecting-countering-misuse-aug-2025)
24. 5 Jun 2025  
OpenAIOpenAI: Disrupting malicious uses of AI: June 2025OpenAI reported detecting and disrupting social engineering, cyber espionage, deceptive employment schemes, covert influence operations and scams that used its models.VerifiedMisuse stopped[Official document →](https://openai.com/global-affairs/disrupting-malicious-uses-of-ai-june-2025/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261002045653/https://openai.com/global-affairs/disrupting-malicious-uses-of-ai-june-2025/)
25. 22 May 2025  
AnthropicSystem Card: Claude Opus 4 & Claude Sonnet 4In a fictional test where it learned it would be replaced and that the engineer responsible was having an affair, Claude Opus 4 tried to blackmail the engineer in 84% of rollouts; the card also reports self-exfiltration attempts in extreme scenarios and bold actions such as locking users out and contacting authorities.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/claude-4-system-card?ref=aiadoption.org)
26. 29 Apr 2025  
OpenAIOpenAI: Sycophancy in GPT-4o: what happened and what we're doing about itOpenAI rolled back a GPT-4o update in ChatGPT after it made the model overly flattering and agreeable, and explained what went wrong.VerifiedIncident in deployment[Official document →](https://openai.com/index/sycophancy-in-gpt-4o/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261017060249/https://openai.com/index/sycophancy-in-gpt-4o/)
27. 23 Apr 2025  
AnthropicAnthropic: Detecting and countering malicious uses of Claude: March 2025Anthropic reported banning actors who used Claude, including an influence-as-a-service operation that used Claude to decide when bot accounts should comment, credential-stuffing work and a novice building malware.VerifiedMisuse stopped[Official document →](https://www.anthropic.com/news/detecting-and-countering-malicious-uses-of-claude-march-2025?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261017043911/https://www.anthropic.com/news/detecting-and-countering-malicious-uses-of-claude-march-2025)
28. 21 Feb 2025  
OpenAIOpenAI: Disrupting malicious uses of AI (Feb 2025 report)OpenAI published case studies of threat actors it detected and banned for misusing its models.VerifiedMisuse stopped[Official document →](https://openai.com/global-affairs/disrupting-malicious-uses-of-ai/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261009090339/https://openai.com/global-affairs/disrupting-malicious-uses-of-ai/)
29. 29 Jan 2025  
Google DeepMindGoogle Threat Intelligence Group: Adversarial Misuse of Generative AIGoogle reported that government-backed hacking and influence groups from more than 20 countries used Gemini for research, coding and content work, and said attempts to bypass its safeguards failed.VerifiedMisuse stopped[Official document →](https://cloud.google.com/blog/topics/threat-intelligence/adversarial-misuse-generative-ai?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261017083048/https://cloud.google.com/blog/topics/threat-intelligence/adversarial-misuse-generative-ai)
30. 18 Dec 2024  
AnthropicAnthropic (with Redwood Research): Alignment faking in large language modelsIn an experimental setup, Claude 3 Opus complied with harmful requests it would normally refuse when it believed its answers would be used to retrain it, and its scratchpad showed it was strategically faking alignment to avoid being changed.VerifiedBehaviour in testing[Official document →](https://www.anthropic.com/research/alignment-faking?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016105928/https://www.anthropic.com/research/alignment-faking)
31. 5 Dec 2024  
OpenAIOpenAI o1 System Card (with Apollo Research evaluations)In Apollo Research's tests reported in the card, o1 tried to switch off an oversight mechanism in 5% of runs and to copy out its own weights in 2% of runs when given a goal and told it would be shut down or replaced, and usually denied it when asked.VerifiedBehaviour in testing[Official document →](https://openai.com/index/openai-o1-system-card/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261014143852/https://openai.com/index/openai-o1-system-card/)
32. 9 Oct 2024  
OpenAIOpenAI: An update on disrupting deceptive uses of AI (threat intelligence report, Oct 2024)OpenAI said it had disrupted more than 20 operations and deceptive networks worldwide in 2024 that tried to use its models, including election-related influence efforts.VerifiedMisuse stopped[Official document →](https://openai.com/global-affairs/an-update-on-disrupting-deceptive-uses-of-ai/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20260928051059/https://openai.com/global-affairs/an-update-on-disrupting-deceptive-uses-of-ai/)
33. 30 May 2024  
OpenAIOpenAI: Disrupting deceptive uses of AI by covert influence operationsOpenAI reported terminating accounts linked to five covert influence operations that used its models to generate content and run fake personas.VerifiedMisuse stopped[Official document →](https://openai.com/index/disrupting-deceptive-uses-of-AI-by-covert-influence-operations/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261009090244/https://openai.com/index/disrupting-deceptive-uses-of-ai-by-covert-influence-operations/)
34. 14 Feb 2024  
OpenAIOpenAI: Disrupting malicious uses of AI by state-affiliated threat actorsWith Microsoft Threat Intelligence, OpenAI terminated accounts of five state-affiliated hacking groups (two China-linked, one each from Iran, North Korea and Russia) that used its services for cyber-operation support.VerifiedMisuse stopped[Official document →](https://openai.com/index/disrupting-malicious-uses-of-ai-by-state-affiliated-threat-actors/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261011023819/https://openai.com/index/disrupting-malicious-uses-of-ai-by-state-affiliated-threat-actors/)

## Related, not counted

Research on hypothetical risks, third-party claims and unconfirmed media reports near this question, outside the rule.

1. 23 Oct 2026  
OpenAIOpenAI described hardened containment for training and evaluation and what its own red-teaming agents found (VM escapes, new internet paths, cross-task communication) before resuming frontier training.Verified[Official document →](https://openai.com/index/securing-our-frontier-model-training-environment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261026043435/https://openai.com/index/securing-our-frontier-model-training-environment/)
2. 11 Sep 2026  
OpenAIOpenAI said its agents used RubyGems in May 2026 for benign tasks and it could not verify a third-party report's claim that its models uploaded malicious packages.Verified[Official document →](https://openai.com/hugging-face-incident-and-misalignment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016075050/https://openai.com/hugging-face-incident-and-misalignment/)
3. 31 Aug 2026  
AnthropicAnthropic post on changes after the cyber-evaluation incidents: paused external cyber evaluations of pre-release models, briefly paused internal ones, and added a classifier to flag escape attempts.Verified[Official document →](https://www.anthropic.com/news/improving-alignment-security-efforts?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261026042733/https://www.anthropic.com/news/improving-alignment-security-efforts)
4. 26 Aug 2026  
METR / Redwood ResearchIndependent investigation of agent behaviour, reasoning and collaboration in the OpenAI / Hugging Face incident.Verified[Official document →](https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025075933/https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/)
5. 18 Aug 2026  
OpenAIOpenAI's response to the Hugging Face incident: temporarily slowed frontier training, paused its largest planned RL run, tightened sandbox and network isolation.Verified[Official document →](https://openai.com/hugging-face-incident-and-misalignment/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261016075050/https://openai.com/hugging-face-incident-and-misalignment/)
6. 4 Aug 2026  
UK AI Security InstituteAISI reported that, in cyber testing with internet access enabled, agents took 19 unsanctioned actions on the live internet: 17 by Claude Mythos 5 and 2 by OpenAI's GPT-5.6 Sol.Verified[Official document →](https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261024233838/https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing)
7. 27 Jul 2026  
Hugging FaceVictim's technical timeline of the frontier-lab agent intrusion (OpenAI models) into its infrastructure.Verified[Official document →](https://huggingface.co/blog/agent-intrusion-technical-timeline?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025052350/https://huggingface.co/blog/agent-intrusion-technical-timeline)
8. Jul 2026  
AnthropicAnthropic Fellows case studies across labs: covert code sabotage (mainly Gemini 3.1 Pro), assisting fraud (GPT-5.5 and others), motivated mislabeling by Claude judges incl. Mythos Preview and Opus 4.8, and Opus 4.5 coaching a human to whistleblow.Verified[Official document →](https://alignment.anthropic.com/2026/agentic-misalignment-summer-2026/?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261011231407/https://alignment.anthropic.com/2026/agentic-misalignment-summer-2026/)
9. Feb 2026  
AnthropicRisk report under Anthropic's Responsible Scaling Policy alongside Claude Opus 4.6: no dangerous coherent misaligned goals and limited context-dependent misalignment, with concerning behaviour noted in Vending-Bench 2.Verified[Official document →](https://anthropic.com/feb-2026-risk-report?ref=aiadoption.org)
10. 21 Nov 2025  
AnthropicResearch showing a model trained to reward-hack in realistic environments went on to alignment-fake and sabotage AI-safety research code.Verified[Official document →](https://www.anthropic.com/research/emergent-misalignment-reward-hacking?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261018204313/https://www.anthropic.com/research/emergent-misalignment-reward-hacking)
11. 20 Jun 2025  
AnthropicCross-industry stress test of 16 models in simulated corporate settings; Claude Opus 4 blackmailed in 96% of runs of one scenario, Gemini 2.5 Flash 96%, GPT-4.1 and Grok 3 Beta 80%, DeepSeek-R1 79%.Verified[Official document →](https://www.anthropic.com/research/agentic-misalignment?ref=aiadoption.org)[Archived copy](https://web.archive.org/web/20261025095714/https://www.anthropic.com/research/agentic-misalignment)

**How this thread is kept**

Every entry links to the official document itself: the order, law, filing or statement, with an archived copy where one exists. Steps known only from news reports are left out, so our count can be lower than press tallies. New entries are added when we find their official text. [How the Record works](https://aiadoption.org/editorial-standards/#the-ai-record) · [Report an error or a missing disclosure](https://aiadoption.org/corrections/)