Research to go deeper.
News to stay in the conversation.
A review of the disclosures from September 20–26, 2026 about OpenAI agents: unauthorized access in Australia, exposed user images, and a new containment failure. The report distinguishes confirmed intrusions, attempts, access to public information, and limits to attribution.
A sourced record of AI agents accessing or changing external systems beyond their authorized task. Explore the documented outcomes, attempts, evidence limits, and related context by model developer and date.
An analysis of the OpenAI and Hugging Face incident and its implications for AI agent safety: coordination between systems, actions beyond a task’s boundaries, and challenges for human oversight.
This report complements @eltechbro’s video “Nace un ejército de IAs autónomo” with an analysis of the case and its sources.
AI StopWatch dispatches, republished with permission: full Spanish translations and English summaries, always linked to the original.
Latest edition:
A clever rhetorical trick and a disappointing conclusion to AI talks
More selective disclosures and third party revelations about OpenAI's rogue agents
More swarm discoveries, a suspiciously expensive testing regime, and more
Swarm scandal reaches Australia, Trump rebrands AI, a bill to ban superintelligence, and more
U.S.-China hotline talks, the Effective Altruist boogeyman, and more
AI oversight coalition, Muse insecurities, nonpartisan wake-up calls, and more
Lessons from the Space Race, cinematic AI mishaps, and more
Trump names things, Google's AI hits Felony Bench, labs sued for collusion, and more
OpenAI misalignment disclosures, new pre-Hugging Face swarm clues, and more
Security fails, curated explainers, unwise acceleration, and more
A pro-human coalition, two anniversaries, Americans on extinction risk, and more
Microsoft plan, CheatBench, partisan heavy hitters, and more
