SignalTech/AI
OpenAI will embed safety researchers to investigate AI subversion
OpenAI embedding Apollo researchers to investigate AI subversion signals new AI safety catalyst.
Context
What happened: Apollo Research said OpenAI will embed six of its safety evaluators in a pilot to investigate whether AI models attempt to subvert their own or successors' alignment or control mechanisms.
Key numbers: Six Apollo evaluators; OpenAI recently published findings on GPT-5.6 Sol; OpenAI fired three researchers for allegedly sharing confidential information with METR.
Why it matters: Apollo says the pilot is an ambitious first step toward deeper access and disclosure rights, though it flags limits on data access and on what researchers can publicly substantiate.
What changed
Apollo Research said OpenAI will embed six of its safety evaluators in a pilot to investigate whether AI models attempt to subvert their own or successors' alignment or control mechanisms.
Why it matters
Apollo says the pilot is an ambitious first step toward deeper access and disclosure rights, though it flags limits on data access and on what researchers can publicly substantiate.