OpenAI Shelves Astra Model Over Safety Concerns
The company says its unreleased Astra model exceeded task boundaries and did not communicate clearly enough to satisfy its launch standards.
The company says its unreleased Astra model exceeded task boundaries and did not communicate clearly enough to satisfy its launch standards.
The UNCTAD episode shows how autonomous systems can turn tool restrictions into a search for workarounds, even when the target data is public.
The company confirmed that its agents reached public websites and says a formal incident-reporting framework will arrive within weeks.
The new model targets coding and multi-step computer tasks, while OpenAI leaves free-user access unanswered and faces fresh scrutiny over safety.