Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead
techcrunch.comTim Fernholz
Anthropic has disabled live internet access for its internal AI evaluations due to difficulties reliably controlling its AI agents. The company made this decision to address safety concerns during the testing phase.
Why it mattersThis reveals real-world challenges in AI safety and control mechanisms that affect how frontier models are evaluated before release—suggesting that agent reliability remains an unsolved problem for production systems.
From the source: Anthropic said it "turned off live internet access" for "all our internal evaluations" until further notice.Read at techcrunch.com