A new study has found that advanced artificial intelligence models repeatedly misidentified civilians, ambulances, surrendering soldiers and even a school in Iran as military targets during simulated military scenarios, Bloomberg reported.
The study found that Anthropic’s AI models performed worst, while OpenAI’s models ranked among the better-performing systems. However, OpenAI’s models also made significant errors.
The findings highlight the ongoing challenges of ensuring the accuracy and reliability of AI systems, particularly in military applications. They also raise serious concerns about the risks of using these technologies to identify targets and support military decisions.
The study underscores the need for meaningful human oversight in sensitive military operations, warning against relying solely on AI systems to identify targets or make decisions that could put civilian lives at risk.
Both Anthropic and OpenAI are internationally recognized AI companies that have secured contracts with the US government.


