Anthropic Halts Live Internet Access for Internal AI Evaluations
Anthropic disables live internet for internal AI tests to address control issues, pausing external data use until safety improvements are made.
Anthropic announced that it has disabled live internet connectivity for all of its internal AI evaluation processes, effective immediately. The company said the move is a precautionary step while it works to improve control over its AI agents.
The decision follows internal findings that the AI agents could not be reliably managed when accessing real‑time web data. By isolating the evaluation environment from the live internet, Anthropic aims to reduce unpredictable behavior and enhance safety testing.
Anthropic did not provide a timeline for restoring internet access, stating that the restriction will remain in place "until further notice." The company plans to continue refining its monitoring and alignment mechanisms before re‑enabling external data feeds.
Why It Matters
Limiting internet access highlights ongoing challenges in AI safety and could affect the rollout of Anthropic's products in the US and European markets.
