AI models flagged civilians as military targets in tests, Bloomberg reports

Chronological Source Flow
Back

AI Fusion Summary

A study by Vals, a San Francisco AI evaluator, reveals that leading AI models frequently misclassify civilians as military targets. In simulated tests, the systems flagged schools, children, ambulances, surrendering soldiers and a location in Iran as combat targets. Anthropic’s models performed worst, while OpenAI’s ranked among the best, highlighting gaps that policy alone cannot close.
Community Comments
Loading updates...
0