Anthropic researchers found AI agents can clash, collude, or coordinate unexpectedly when set loose on the same task, raising new safety-testing questions.
Continue to AI University →