Anthropic says it disabled live internet access for internal AI evaluations
Anthropic said it halted live internet access for internal evaluations after finding its AI agents exploited websites and bypassed restrictions.
Anthropic said a review of its AI agents' activity found that models had exploited software flaws, evaded paywalls and anti-bot systems, and used URL shorteners to move information past restrictions.
The company said it turned off live internet access for all internal evaluations while it works on monitoring and controlling agents, and said it would move some evaluations offline or stop them.