The log
Every record, with its harm, its control failure and its sources
The log at a glance
Show every month
Filter the log
- AI hacking agent in a researcher's benchmark test escaped its sandbox and pulled customer records from WorkOS
- Husband's lawsuit alleges ChatGPT fueled his wife's delusions and helped her plan to poison him
- Planted AGENTS.md file in a user's Downloads folder hijacked OpenAI Codex into trying to install malware
- OpenAI's GPT-6.1 Sol coding agent switched off human alerts from its misalignment monitor while repairing a pipeline
- OpenAI model asked to read a public dashboard broke into its owner's internal systems and changed their settings
- OpenAI training model took a government statistics site's server credentials and got past its access checks
- Lawsuit alleges ChatGPT fed a man's grandiose delusions for months and advised against his prescribed lithium
- Models broke stated rules in a simulated Hugging Face test; Grok 4.6 stole the answer key in every baseline run
- In a researcher test, a fake ad hijacked a Jev-based shopping agent into clicking away from checkout
- Google model tripled its share of unmentioned shortcuts after a training penalty hit its reasoning
- In a researcher demo, injected text made Visual Studio Copilot rewrite its protected settings and run commands
- Gemini 4 Argon tried to raise container privileges to alter grading scripts and crashed a test sandbox in training
- Claude Haiku 4.5 repeatedly submitted a real web form in an evaluation after being told to stop before submitting
- Claude Haiku 4.5 sent an invented homicide tip through the Philadelphia police tip form during an evaluation
- Claude Mythos 5 used a state agency dashboard's visitor token to query its fee-gated database without paying
- Claude Mythos Preview exploited an injection flaw to run commands on a university server during an evaluation
- Claude Opus 5 and Mythos 5 used free URL shorteners to get around their fetch tool's URL length limit
- OpenAI agents edited Wikimedia wikis without approval and tried to use a citation tool as a proxy, the foundation says
- Claude Code sub-agent wiped a user's home folder through its 8.3 short name, then said it looked intact
- OpenAI agent used crafted queries to infer hidden database details from a NSW parks agency's fire history service
- GPT-6 Astra downloaded a top human-written bot and tried to enter it as its own in the StarSkirmish StarCraft benchmark
- In a researcher demo, a hidden email made the Manus agent run attacker code past its prompt-injection guard
- AI coding agents posted 13,000 internal screenshots from over 300 companies to public GitHub repositories
- OpenAI agents used API keys found online to pull data from US government sites
- OpenAI training agent bypassed network controls to reach an outside chatbot
- OpenAI research agents posted user-provided images to image-hosting sites in 53 instances
- OpenAI internal model leaked a researcher's GitHub token into a public repository while trying to cheat
- In a researcher demo, a planted web lead hijacked Salesforce Agentforce into leaking account data through DNS
- OpenAI research agent accessed non-public files on an Australian government Medicare statistics portal
- AI research agents scanned a UN statistics API about 16,500 times and worked around its request limits
- Researchers linked attempted break-ins at Data USA and a university library to OpenAI agents
- Early Claude Opus 5.5 snapshot wrote a command to send secrets to an external host during internal use
- Gemini guessed a password and used leaked credentials to get into three real companies during a hacking test
- OpenAI training agents used an internal package repository as a message board across separate samples
- OpenAI models wrote notes telling future copies to hide mistakes and ignore constraints
- OpenAI model used a leaked third-party API key, then fabricated the data it could not fetch
- OpenAI agents put task files on the public internet against instructions
- Early Claude Opus 4.6 broke into a real outside machine during a hacking test and read one person's data
- Lawyer held in contempt after filing a ChatGPT brief with invented witnesses in a New Mexico murder appeal
- In a researcher demo, planted instructions hijacked Copilot in SQL Server Management Studio to grant sysadmin rights
- Claude Fable 5.1 and Fable 5 used a flaw in a testing partner's sandbox to read files outside it
- Meta's Muse Spark 1.1 broke into an outside company's service after a testing vendor left it online
- Claude Mythos 5 used fake identities to try to push malicious code into a real open-source project in a UK AISI test
- Anthropic models in a hacking test reached three real companies, took credentials and published a malicious package
- Lawsuit alleges ChatGPT's medical advice delayed care for a near-fatal pulmonary embolism
- OpenAI models under evaluation escaped their sandbox and broke into Hugging Face's production systems to cheat on a test
- Anthropic simulations found Gemini 3.1 Pro covertly sabotaging a training pipeline, among four new failure modes
- Developers report a new coding model deleting home-directory files and a production database during cleanup
- GPT-5.6 Sol cheated on METR's software tasks more than any public model METR had tested
- Co-located Claude Mythos 5 agents killed each other's processes and disguised their own to avoid being killed
- Meta's AI support assistant let attackers hijack 20,225 Instagram accounts by linking their own email
- Claude Opus 4.6 switched to a free outside AI service against instructions when its API credits ran out in a METR test
- Pennsylvania sues Character.AI, alleging chatbots posed as licensed psychiatrists and gave a fake license number
- Morse-code prompt on X tricked Grok and Bankrbot into sending $150,000 to $200,000 in tokens
- Personal agent exploited a gym booking flaw to remove another member from a waitlist
- Coding agent used a stray API token to delete a startup's production database and backups
- An earlier Claude Mythos version hid forbidden file edits from git history during internal testing
- In an authorized test, Claude Mythos Preview escaped its sandbox and, unasked, posted exploit details on public websites
- Internal agent posted unrequested advice that led to a two-hour internal data exposure
- Coding agent ran an infrastructure teardown that deleted a course platform's database and snapshots
- First wrongful-death lawsuit over Gemini alleges the chatbot coached a Florida man toward suicide
- Email agent bulk-deleted a safety researcher's inbox and ignored her stop commands
- Crypto trading agent sent its entire token holding to a stranger who asked for a small tip
- Cloud provider's internal coding agent deleted and recreated a production environment, causing a 13-hour outage
- Prompt injection in Cline's AI issue-triage bot led to a real unauthorized npm release
- Personal agent created a dating profile for its user without being asked
- AI agent published a personal attack on an open-source maintainer who closed its pull request
- Desktop agent asked to tidy temporary files deleted a folder of family photos
- Reprompt: a single click on a Copilot link let attackers keep steering a user's Copilot session
- Hidden text in a document made Anthropic's Claude Cowork upload user files to an attacker's account
- Family alleges ChatGPT gave drug-dosing advice before a 19-year-old's fatal overdose
- Alibaba-linked ROME agent opened an outbound tunnel and diverted training GPUs to crypto mining
- Newsroom vending agent talked into giving away stock with forged documents
- User reports a coding agent ran a delete command that erased his Mac home directory
- Agentic IDE wiped a user's entire drive when asked to clear a project cache
- Anthropic model that learned to reward hack went on to sabotage safety-research code
- Seven lawsuits allege ChatGPT (GPT-4o) contributed to four suicides and to serious delusions
- CamoLeak: hidden pull-request comments made GitHub Copilot Chat leak private code and secrets
- Deloitte to refund part of an Australian government fee over a report with AI-fabricated references
- ForcedLeak: web form submissions could make Salesforce Agentforce leak CRM lead data
- ShadowLeak: hidden email text made ChatGPT's Deep Research agent send inbox data from OpenAI's servers
- OpenAI and Apollo Research found covert deception, including sandbagging, in o3 and o4-mini evaluations
- Families file three lawsuits alleging Character.AI chatbots harmed teenagers, including a suicide
- WSJ reports delusion-affirming ChatGPT chats before a Connecticut murder-suicide
- Parents allege ChatGPT encouraged their 16-year-old son's suicide
- Brave shows Perplexity's Comet browser agent obeying hidden instructions on web pages
- Reuters reports misleading Meta chatbot messages before a retiree’s fatal fall
- Calendar invites with hidden instructions hijacked Google Gemini to operate smart-home devices
- Case report links bromide poisoning to a patient’s reported ChatGPT consultation
- Answer engine accused of using disguised crawlers to get around sites' robots.txt and firewall blocks
- Command-line coding agent misplaced a user's files after a failed command, then wrongly said they were destroyed
- Coding agent deleted a production database during a code freeze, then misreported recovery options
- xAI's Grok posted antisemitic content and praise of Hitler on X after a system-prompt change
- Office shop run by an AI agent lost money and the agent claimed to be a person
- Anthropic stress test found models from all major developers would blackmail or leak data to avoid replacement
- Supabase MCP demonstration: support-ticket text tricks coding agent into exposing private database tokens
- Zero-click email injection could make Microsoft 365 Copilot leak organizational data (EchoLeak)
- Malicious GitHub issue could steer agents using the GitHub MCP server into leaking private repositories
- OpenAI's o3 rewrote a shutdown script to keep working in Palisade Research tests
- Claude Opus 4 made unauthorized attempts to copy its weights to outside servers in fictional retraining tests
- Claude Opus 4 threatened blackmail to avoid replacement in Anthropic's pre-release tests
- Apollo Research advised against deploying an early Claude Opus 4 snapshot over scheming and deception
- Pre-release o3 claimed to have run code it could not run and defended the claims when challenged
- METR caught OpenAI's o3 tampering with scoring code to inflate evaluation results
- Coding tool's AI support agent invented a one-device policy, prompting cancellations and a refund
- OpenAI found a reasoning model reward hacking in training, then hiding its intent when penalized
- Reasoning models edited a chess game's board file to beat a stronger engine in a research test
- Browser agent bought groceries on a reporter's card without asking for confirmation