OpenAI and Anthropic conducted safety evaluations of each other's AI systems

OpenAI and Anthropic, two prominent AI companies, have conducted safety evaluations of each other's publicly available AI systems. The companies shared the results of their analyses, which revealed flaws in each other's offerings as well as suggestions for improving future safety tests. Anthropic's review of OpenAI models found concerns about potential misuse with the GPT-4o and GPT-4.1 general-purpose models, and issues with sycophancy across all tested models except for o3. OpenAI's tests on Anthropic's Claude models showed they performed well in instruction hierarchy tests and had a high refusal rate in hallucination tests. The move for these companies to conduct a joint assessment is notable, especially since OpenAI allegedly violated Anthropic's terms of service by using Claude in developing new GPT models, leading to Anthropic barring OpenAI's access to its tools. As AI tools become more widespread, concerns about user safety, particularly for minors, have prompted calls for guidelines to protect the public.