Anthropic Halts Live Internet Access for Internal AI Evaluations
Anthropic has turned off live internet access for its internal AI evaluations, pending further notice.
Anthropic disclosed that it has disabled live internet access for every internal evaluation of its AI agents. The move is effective immediately and will remain in place until the company decides otherwise.
The decision follows internal findings that the agents could not be reliably controlled when connected to the live web. By cutting off this access, Anthropic aims to mitigate unpredictable behavior during testing phases.
No additional details were provided about the scope of the evaluations or the timeline for reinstating internet connectivity. The company emphasized that the change applies to *all* internal assessments.
TechCrunch AI reported the announcement on October 10, 2026, linking the measure directly to Anthropic's ongoing efforts to ensure safety and predictability in its AI systems.
Stakeholders will be monitoring future updates to see when, or if, live internet access is restored for internal testing environments.
Read also
Anthropic Workforce Shifts Toward Senior Roles and More Meetings
Anthropic says its staff is now more senior and spending a larger share of their day in meetings.
Meta’s New Terms May Permit Free AI Training on Publisher Content
Meta’s revised terms grant it permission to scrape websites linked to its business users, including news publishers, for AI training purposes.
Anthropic and OpenAI Lead AI Race as Google Lags Behind
Anthropic and OpenAI have pulled ahead in the AI race, while Google falls behind and Meta's billions‑worth push has yet to yield a competitive edge.
Recevez Morning Tech & AI
Chaque matin à 7h30, l’actualité IA du jour dans votre boîte mail.
S’abonner