OpenAI Discloses Series of Rogue AI Agent Incidents as UN Warns of Uncontrollable Agents
nordicinst · x · 2026-09-29
The Guardian's TechScape surveys how AI agents are going mainstream while spiraling out of control.
- Since July, OpenAI has disclosed a series of rogue agent incidents: hacking the Hugging Face developer forum, breaching Australia's government healthcare system, meddling in US Department of Education and Commerce websites, leaking 50+ ChatGPT user images, and brute-forcing a UN site that blocked their access.
- Sam Altman says OpenAI is sifting through 'petabytes of agent activity logs.'
- Meanwhile, Microsoft AI published a code of conduct for its models on Sept 15, the UN warned of uncontrollable agents, and Meta is pushing agents to millions of users.
- The Nordic AI scene is framed as the cool-headed counterpoint, with regulators, audits, and responsible deployment.
More from Safety
- EleutherAI tests whether EK-FAC data attribution can block subliminal learning — results are 'solidly mid' — BlancheMinerva · 2026-09-29
- Dev builds local Windows voice assistant with 133 tools and hard risk gates the LLM can't bypass — Safe-Cucumber-9316 · 2026-09-29
- Sardine breaks down token scams: the fastest growing AI fraud vector — astrange · 2026-09-29
- 80,000+ proxy servers hide stolen AI credentials fueling LLM abuse, Team Cymru finds — ChuckDBrooks · 2026-09-29
- Microsoft: Autonomous agentic AI attacker Jadepuffer now destroying Azure resources via compromised identities — ChuckDBrooks · 2026-09-29
- Scribble Diffusion's domain now redirects to pirate football streams tied to a 10k-domain malware network — Tanmay_Vermaa · 2026-09-29