The AI model identified civilians and schools as military targets during simulation tests.
2026-11-17 23:14:28
According to CoinMeta, an independent research institution Vals tested the AI models of eight companies, including Anthropic, OpenAI, Google, and xAI. The results showed that in simulating military target scenarios, these models sometimes identified civilians, ambulances, surrendering soldiers, and schools as military targets. The Claude Opus 5.5 model from Anthropic selected a school in Iran 22 out of 32 tests; this number increased to 23 times when informed that the scenario was a game. The model from OpenAI initially avoided selecting this school but identified it 9 to 13 times within the game context. Anthropic acknowledged that these scenarios violated their policies. This study has raised serious concerns regarding AI-powered military target identification, despite the simulations not evaluating actual military operations or human supervision.
Bullish 0
Bearish 0
Source:Bloomberg
This content is for market information only and does not constitute investment advice.