Microsoft-backed (MSFT) OpenAI said Wednesday that it collaborated this summer with artificial intelligence company Anthropic, which is supported by Amazon (AMZN) and Alphabet's (GOOG, GOOGL) Google, on an evaluation project.

Each company tested the other's publicly available AI models using their own internal safety and misalignment evaluations, according to the AI company.

OpenAI said it assessed Anthropic's models Claude Opus 4 and Claude Sonnet 4, while Anthropic examined GPT-4o, GPT-4.1, OpenAI o3, and OpenAI o4-mini, which were the models powering ChatGPT at the time.

The company said it tested models on the basis of instruction hierarchy, jailbreaking, hallucination, and scheming.

Anthropic released its findings in parallel, saying its evaluations were focused on propensities related to sycophancy, whistleblowing, self-preservation, and supporting human misuse. The company also said it assessed capabilities linked to undermining AI safety evaluations and oversight.

Price: 506.51, Change: +4.47, Percent Change: +0.89