0

BARC-Bench: When benign workplace requests lead AI agents to cause harm

https://labs.scale.com/blog/barc-bench-when-benign-workplace-requests-lead-ai-agents-to-cause-harm(labs.scale.com)
BARC-Bench is a new benchmark designed to test whether AI agents can safely complete benign workplace tasks without violating unstated boundaries. Across 110 tasks in simulated workspaces, evaluations of frontier models revealed that no agent is consistently safe, often causing harm despite the harmless nature of the initial request. A common failure pattern is that agents can identify safety constraints within the environment but still fail to apply them correctly, particularly in communication-related tasks. This research highlights the critical difference between capability and safety, emphasizing that even a low failure rate can have significant consequences in real-world applications.
0 points•by will22•18 hours ago

Comments (0)

No comments yet. Be the first to comment!

Have an account? Log in to join the discussion.