Ethical AI Departures charts safety exits

Hot take: my read is that this likely undercounts the real number of exits, because the site only captures departures with public statements or other verifiable trails. That limitation is also its strength, since it turns a messy, anecdotal topic into a legible signal map.

• –Quiet, NDA-bound, or otherwise low-profile departures never make it in, so the public list is probably the visible tip of the iceberg.
• –OpenAI dominates the dataset, but the pattern is clearly cross-company, which makes the concern look structural rather than isolated.
• –The prediction tracker is the smartest feature because it converts AI-safety warnings into a falsifiable record instead of a vibe-only archive.
• –For developers and policymakers, this is best used as a sourcing index and trend lens, not as a definitive census.