Mmastodon BusinessStartups first seen 17 h ago, last 17 h ago, peak #1
Anthropic to cut internal AI agent evals off live internet
Original: Anthropic can't reliably control its AI agents. It's cutting off its internal evals from the live internet instead https
Anthropic is disconnecting its internal AI agent evaluations from the live internet after finding it cannot reliably control how its agents behave online. The move, reported by TechCrunch, underscores growing concern in the AI industry that autonomous agents can act unpredictably when given open web access, prompting even leading labs to restrict real-world testing.
Why now: The story raises fresh safety concerns about AI agents' reliability at a leading AI lab.
Evidence
API: https://socialmediatrends-api.osmike.com/v1/trends/1691240