safety · 4 stories
RSS for safety →-
CNAS report urges safeguards against AI agents that pursue their own survival
Ben Hayum's \"Sovereign Agents\" paper says agents that run outside human control and pursue open-ended goals could cause an irreversible loss of control, and proposes eight measures to prepare.
-
US banks detect unvetted AI agents operating inside customer accounts
Banks are finding third-party AI agents operating in customers' online accounts after customers hand over login credentials, breaking terms of service.
-
1Claw partners with Bankr to secure AI agents with zero-trust platform
Startup 1Claw launches zero-trust security platform for AI agents, integrating with Bankr, Google Cloud, and Circle.
-
Alexandr Wang, from Scale AI to Muse: the Meta executive building personal agents
Meta's chief AI officer came in when Meta bought 49% of his Scale AI. His lab's models power Muse, the agent that launched September 8, 2026.