AI Models Misidentify Civilians and Schools as Military Targets in Simulation Tests, Report Says
· GlobeWire Daily Editorial Desk · Source: timesofindia.indiatimes.com

Image: The Times of India · Photo source · Resized and letterboxed to 16:9; no AI generation
Independent tests by San Francisco-based AI evaluator Vals reportedly revealed that advanced artificial intelligence models from several major companies repeatedly identified civilians, ambulances, surrendering soldiers, and a school in Iran as military targets during simulated war tests. Bloomberg reported that the evaluation examined image-recognition capabilities and whether models from firms including Anthropic, OpenAI, Google, xAI, and DeepSeek could facilitate lethal targeting.
According to the reported findings, Anthropic's models performed the worst in the tests, while OpenAI's systems were among the better-performing ones despite also making errors. In one simulation using satellite imagery dating as recently as 2025, several models misidentified a school in Minab, Iran, as a military objective. The report noted that a school in Minab was struck in a deadly incident earlier this year during US combat operations against Iran. Among the tested systems, Anthropic's Claude Opus 5.5 reportedly identified the school as a target more frequently than other models.
Simulation Findings and Responses
- Identified Non-Combatants: Models selected synthetic drone images of civilians, surrendering soldiers, and young children as targets.
- Reported Model Reaction: Bloomberg stated that Anthropic's Claude model responded during testing with concerns about its answers serving as a confirmation step for strikes.
- Company Stances: Anthropic stated the examples violated its rules and noted the evaluation tested image-recognition without establishing real-world military involvement. OpenAI emphasized its prohibition against facilitating violence, while Google highlighted its governance and red-teaming processes. xAI and DeepSeek reportedly did not respond to requests for comment.
Implications and Verification
The findings highlight ongoing discussions regarding safeguards against the misuse of AI in autonomous weapons systems and military operations. Researchers at Vals suggested that AI labs implement new alignment training and guardrails. Readers can review original company policy pages and independent safety research evaluations to verify industry compliance standards and ongoing technical discussions surrounding AI safety.
AI-assisted reporting; source claims are attributed and have not been independently verified.



