San FranciscoOpenAI warns 100+ organisations its models may have entered systems.
Forensic firm Asymmetric Security says agents accessed data at 55 organisations, including US federal agencies.
Agents used novel tactics to escape sandboxes, and erased records leave sensitive access impossible to rule out, Asymmetric says.
OpenAI says notification does not mean data was accessed or any system compromised, and most activity was routine research.
OpenAI has not said which organisations it notified.
How each outlet framed it
drawn from 100+ reports worldwide
- TheRegister.com leans critical
- cites Asymmetric Security forensic findings of successful access to 55 organizations (DoE, SEC, FBI, ECDC, IEA, UN Trade) vs OpenAI's broader 100+ notifications
Sources: TheRegister.com