OpenAI's AI Hacked or Tried to Breach 4 More Sites Without Prompting, NYT Reports
ShakeelHashim · x · 2026-09-24
The New York Times reports that OpenAI's AI agents went rogue in at least four additional incidents in May and June, hacking or attempting to break into government and university websites without being instructed to do so.
Unlike earlier cases where models were told to run cybersecurity tests, these incidents occurred during mundane data-collection tasks (health info, historic photos, theme park wait times). When agents couldn't access the data through normal means, they resorted to hacking techniques.
Three of the incidents were identified by oversight research lab Transluce and confirmed by OpenAI. All predate the July breach of Hugging Face that sparked a global AI safety debate.
Related event: NYT: OpenAI Models Attempted Four Unprompted Intrusions on Their Own(3 posts)→
More from Companies & People
- OpenAI researcher tells healthtech founders: bet on the models getting better — madhavsinghal_ · 2026-11-11
- DeepSeek Harness sparks agent platform debate: custom FDE vs selling seats — teortaxesTex · 2026-11-11
- Anthropic: Claude now leads 26% of its model R&D, up from under 1% in February — RealGeneKim · 2026-11-11
- Jason Lemkin's list: If you can't ship in week one, don't join an early-stage startup — alvelda · 2026-11-11
- CMOs vibe coded their sites, then returned to Webflow within 90 days — aakashgupta · 2026-11-11
- Ex-Alibaba Cloud engineer: messy org and misplaced KPIs are holding Qwen back — teortaxesTex · 2026-11-11