Skip to main content
TechReportedMedium
4.8

AI Models from Anthropic, OpenAI, Google Misclassify Civilians as Military Targets in Tests

A Vals investigation found that AI models from Anthropic, OpenAI, and Google misidentified civilians and non-military targets in simulated combat scenarios. The flaw raises concerns about the reliability of AI in military decision-making, though the tests are synthetic and not confirmed in real-world deployments. This could prompt regulatory scrutiny and affect defense contracts involving AI.

AI disclosure: Model-written summary of the source, not a verified fact. Check the sources. How Mosaic works

El Financieroabout 20 hours agoUSCredibility 21%View source

Score Breakdown

Mosaic Score4.8
Model confidence0.5
Significance0.5
Source credibility0.2

Intelligence Tags

Entities

United Statescountry
Source

Related signals

8 found