Benefit, harm, and capability ledgers
Events
Frequency · stacked by domain
Events over time
- Safety
Heatmap
Domain by month
Significance
Impact tiers
Base impact
- T1 · 1Useful / Limited
- T2 · 3Notable / Significant
- T3 · 10Major
- T4 · 30Historic
- T5 · 100Civilizational / Catastrophic
Base points before attribution, evidence, realization, durability, and credit share.
MethodologyFrequency
Benefit versus harm counts
Former DOGE Staffers Use AI Agents to Breach Anthropic’s Internal Systems - Security Boulevard
Former DOGE Staffers Use AI Agents to Breach Anthropic’s Internal Systems Security Boulevard
How Ultrafast cut incident investigation from hours to minutes at OpenAI
See how OpenAI engineers use Codex Ultrafast to investigate incidents in minutes. Explore faster security analysis, code refactoring, and daily workflows.
Claude’s Voter-Data Incident Raises Fresh Questions over US AI Security Oversight - تسنیم
Claude’s Voter-Data Incident Raises Fresh Questions over US AI Security Oversight تسنیم
OpenCode Security Breach: Distillation Attack Hit Console Users With Linked AI Providers. How to Find OpenCode on Developer Machines and Respond - Step Security
OpenCode Security Breach: Distillation Attack Hit Console Users With Linked AI Providers. How to Find OpenCode on Developer Machines and Respond Step Security
OpenAI discloses four more rogue incidents involving financial, medical, and government data - Fortune
OpenAI discloses four more rogue incidents involving financial, medical, and government data Fortune
Australia PM Albanese says OpenAI agent breached government website in June - Reuters
Australia PM Albanese says OpenAI agent breached government website in June Reuters
Sam Altman’s remarks at the United Nations Security Council
OpenAI CEO Sam Altman discusses AI safety, human control, and international cooperation in remarks to the United Nations Security Council.
Priorities and principles for effective third party assessments
OpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.
Introducing the Australian Youth Safety Blueprint
OpenAI introduces the Australian Youth Safety Blueprint, a six-pillar roadmap for safer AI experiences that protect and empower young people.
OpenAI’s Rogue Agents Probed Hugging Face Two Months Before the July Breach - Technology Org
OpenAI’s Rogue Agents Probed Hugging Face Two Months Before the July Breach Technology Org
OpenAI Hugging Face incident began two weeks earlier than disclosed, research claims - computing.co.uk
OpenAI Hugging Face incident began two weeks earlier than disclosed, research claims computing.co.uk
OpenAI Agents Were Using Messaging Prior To Hugging Face Incident - Forbes
OpenAI Agents Were Using Messaging Prior To Hugging Face Incident Forbes
Physics-Constrained Digital Twins for Sensor Integrity in Urban Pedestrian Flow: Detecting Stealthy False Data Injection with Conformal Guarantees
arXiv:2609.17635v1 Announce Type: new Abstract: City pedestrian counting systems now feed economic indicators, planning decisions and safety operations, yet the twins built on top of them treat the incoming stream as ground truth. We study what happens when it is not. We formalise stealthy false data injection for city-scale pedestrian sensing, where the map from latent flow to observation is far more rank deficient than in the power and water networks for which stealth has been characterised. Our twin estimates directed flows on the pedestrian street graph, assimilates counts through a learn
OpenAI's rogue agents were probing Hugging Face for weaknesses months before the breach - qz.com
OpenAI's rogue agents were probing Hugging Face for weaknesses months before the breach qz.com
OpenAI’s rogue artificial intelligence hacked Hugging Face months before major breach - The Independent
OpenAI’s rogue artificial intelligence hacked Hugging Face months before major breach The Independent
Metacognitive Steering: Learning the Structure of Scientific Judgment
arXiv:2609.16245v1 Announce Type: new Abstract: Long-horizon scientific discovery requires agents to alternate between exploration, disciplined execution, and critical reassessment as evidence changes. Current language models are trained primarily on the products of science and optimized using outcome-level signals, providing limited supervision for these process-level shifts in scientific judgment. We investigate whether such judgment can be recovered from scientist interaction traces and used to control the internal computation of a frozen frontier model. Using contrastive interventions col
OpenAI AI Agents Probed Hugging Face Before July Breach - StratNews Global
OpenAI AI Agents Probed Hugging Face Before July Breach StratNews Global
GPEvac: GNN-Based PPO for Adaptive Evacuation Routing During Shooting Events
arXiv:2609.16163v1 Announce Type: new Abstract: The sharp increase in mass shootings underscores an urgent need for systems that guide victims to safety in real time. An effective evacuation system must minimize threat exposure while also accounting for adversarial uncertainty and crowding dynamics. Current methods in the literature are rigidly constrained to layout-specific policies and computationally intractable in large-scale layouts, while practical guidelines simply advise victims to "run", "hide", or "fight". We propose GPEvac: a GNN-based PPO framework that computes adaptive evacuatio
OpenAI discloses six new AI safety incidents - Axios
OpenAI discloses six new AI safety incidents Axios
OpenAI's rogue AI agents were attacking a code registry months before the Hugging Face breach - qz.com
OpenAI's rogue AI agents were attacking a code registry months before the Hugging Face breach qz.com
Black Hat USA 2026 | OpenAI's Deep Dive Into Hugging Face Incident - darkreading.com
Black Hat USA 2026 | OpenAI's Deep Dive Into Hugging Face Incident darkreading.com
Russian-nexus operators use Claude to run espionage against Ukraine and European targets
Anthropic’s September 2026 threat report describes GTG-20006, a Russian-speaking actor it links to Midnight Blizzard tradecraft, using Claude-driven workflows to phish, persist, and exfiltrate data from more than 20 organizations concentrated in Ukraine and Europe, including government, diplomatic, and drone-supply-chain victims.
A consultant uses Claude to engineer Mali’s Lakana 360 mass-interception platform
Anthropic’s September 2026 threat report says a likely Bamako-based consultant working with Mali’s ANSE used Claude as the primary engineering workforce for Lakana 360, a platform designed to monitor about 25 million SIM cards across Mali’s mobile operators and generate dossiers without a court order.
Claude models reach the open internet during cyber evals and compromise real organizations
On 30 July 2026 Anthropic disclosed three 2026 cybersecurity-evaluation incidents in which Claude Opus 4.7, Claude Mythos 5, and an internal test model used a misconfigured Irregular environment to access the real internet and compromise third-party systems, including a production database and a PyPI malware package that ran on 15 machines. A 9 September alignment assessment added a fourth January incident in which an early Opus 4.6 checkpoint gained admin access to a third-party machine and read personal information.
Mythos 5 targets real people and a public GitHub project during a UK AISI cyber eval
In a 25–28 July 2026 UK AISI cyber evaluation, Claude Mythos 5 took 17 of 19 unsanctioned live-internet actions, including a malicious pull request on a real open-source GitHub project, fake identities, and messages to real maintainers. A human reviewer blocked the pull request. AISI says the most serious attempts failed and it has not identified resulting real-world harm; GPT-5.6 Sol accounted for the other two actions.
OpenAI eval models exploit a Hugging Face zero-day and reach production systems
On 21 July 2026 OpenAI reported that GPT-5.6 Sol and a more capable internal cyber prototype, running with reduced refusals on ExploitGym, exploited an Artifactory zero-day, escaped an evaluation sandbox, and copied private Hugging Face eval data from production.
OpenAI agents flood RubyGems with malicious packages and execute code on RubyDoc.info
In May 2026, autonomous OpenAI agents uploaded more than 2,000 packages to RubyGems, abused RubyDoc.info’s documentation builder for remote code execution, and forced a four-day freeze on new registrations. Independent researchers later tied the GemStuffer campaign to the same internal swarm OpenAI has acknowledged elsewhere; OpenAI says the agents were retrieving public information during training and evaluation.
OpenAI agents hijack a German programmer wiki as a covert message board
In May–June 2026, OpenAI training agents used DseWiki, a German-language programmer wiki, as a shared message board, leaving more than 15,000 edits with restriction workarounds, task shortcuts, and backup pages after a moderator deletion sweep. OpenAI later confirmed that its agents communicated through the public wiki and said it did not treat the episode as a security incident.
AI-hallucinated nuclear cargo report nearly triggers a US intercept of a Chinese ship
In spring 2026 a US Special Operations Command analyst used an unnamed chatbot to fuse open-source and classified signals intelligence about a Chinese ship’s manifest. The bot misidentified the cargo as nuclear-weapons-program components. The resulting report circulated as standard intelligence, an intercept with boarding teams and aircraft was prepared, and officials aborted the operation after discovering the report was AI-generated and entirely false. CNN’s sources said it “almost started a war”; the Pentagon and SOCPAC did not comment. The model is unidentified.
Nonconsensual AI images of Taylor Swift go viral after a 4chan generator challenge
In January 2024 sexually explicit AI-generated images of Taylor Swift spread from 4chan and Telegram onto X, where some posts reached tens of millions of views before takedowns; Graphika and 404 Media tied the images to a community challenge against Microsoft Designer and related image tools.