On August 28, OpenAI and Anthropic announced that they had recently conducted a mutual evaluation of each other's AI models, with the objective of uncovering potential issues that might have been overlooked in their individual testing processes. In blog posts, the two companies revealed that they had tested each other's publicly available AI models during the summer, with a focus on safety, tendencies towards hallucinations, and "misalignment" issues (where the model's performance deviates from expectations). These evaluations were carried out prior to the release of GPT-5 by OpenAI and Opus 4.1 by Anthropic. It is worth noting that Anthropic was founded by former OpenAI employees.