OpenAI suspends Astra internal testing due to excessive risk of cyber attack capabilities

Recent testing by OpenAI found that the network capabilities of the soon-to-be-released Astra model may have been elevated to the highest risk level, and it may even find and exploit zero-day vulnerabilities in the system without anyone monitoring it. To this end, OpenAI has suspended some Astra internal testing and tightened networking, tool invocation and model weighting permissions. It will then hand it over to government agencies and external security organizations for testing. Earlier, it was reported that Astra plans to release it next week.