OpenAI's former safety lead: new capability and risk ship 'every Tuesday'

David Robinson, who led safety reporting for OpenAI's frontier models before resigning on Oct. 3, said in his first interview since leaving that reasoning training and new tool integrations now change model capabilities between major releases. He warned that safety testing cannot keep pace and argued for a live dashboard instead of lengthy PDF system cards.

• Robinson resigned Oct. 3; CEO Sam Altman responded publicly on X
• Reasoning training layered on base models sped up releases: GPT-6.1 Sol shipped a week after GPT-6 Sol
• OpenAI research teams now use over 100x more agentic compute than at the start of the year
• Median researcher spends over $600 a day on inference at API prices