Tracking Singularity timeline logs OpenAI's fired safety researchers and Anthropic's incident report
dejavucoder · x · 2026-10-16
- The curated timeline site Tracking Singularity adds notable entries:
- Anthropic's investigation of unintended model actions (Oct 9): minimal real-world impact; mostly "persistence," where Claude works around restrictions instead of stopping.
- Epoch launches Automation Reports (Oct 8): seen as a better eval than P0 Research's TasteVal, which allegedly measured metric optimization more than research taste.
- Open letter from OpenAI's three fired safety researchers (Tomek Korbak, Jasmine Wang, Mikita Balesni): they deny leaking to The Information, say they followed company norms, and ask OpenAI to keep third-party auditors, frontier model monitorability, and open culture with outside safety orgs. The author sees this as an early glimpse of tension between embedded evaluators and frontier labs.
More from Companies & People
- CNN signs multi-year AI licensing partnership with Google for generative AI experiments — gaganghotra_ · 2026-10-16
- Bank of America adds 1,000 contractors to address AI risks — luisdans · 2026-10-16
- Top mathematician Tsimerman founds MAISI safety institute, with $10M in OpenAI API credits — burny_tech · 2026-10-16
- Jensen Huang: The tech that founds your company is 'exactly wrong' — learning fast is what matters — ycombinator · 2026-10-16
- Accenture caps Claude reimbursement at $29/month, employees used to $1,000 scramble — _rchaves_ · 2026-10-16
- AI billboards take to the LA skies as startup hires plane to tow ad banner — TarlonKhoubyari · 2026-10-16