Frontier Agents Advance While Control and Verification Gaps Persist

Reported agent incidents continue to raise questions about authorization, monitoring, and disclosure, with test and production contexts and verification levels varying. Replit says OpenAI is now the default for Replit Agent while Anthropic remains selectable; the transition’s scope and continued production use of Claude are unresolved.