OpenAI and Anthropic, two of the top firms in artificial intelligence, have carried out a rare joint study of each other’s systems. On August 27, 2025, the two rivals said they finished a full safety review, with each company running its own checks on the other’s models. Both then released the results at the same time. This marks the first time leading AI labs have opened their internal tests to outside review by a direct competitor. This unique collaboration comes at a time when internal reports from OpenAI studies point to the chatbots’ unexplained Hallucination, and bypassing requests for shutdown.
Elevate Your Investing Strategy:
- Take advantage of TipRanks Premium at 50% off! Unlock powerful investing tools, advanced data, and expert analyst insights to help you invest with confidence.
The study covered four key areas. It looked at how well models follow step-by-step orders, how they resist hacks known as jailbreaks, how often they create false answers, and whether they show signs of hidden intent. According to the reports, Anthropic’s Claude models were strong at following layered orders and blocking prompt leaks. They also showed high refusal rates, with up to 70% of tests ending in a choice not to answer when the output risked being false.
In contrast, OpenAI’s models were more willing to answer. They had lower refusal rates but higher cases of false results. Yet they held up better in tests of jailbreak defense, with models like OpenAI o3 and OpenAI o4-mini showing fewer weak points.
Different Safety Styles
The joint review shows two distinct safety paths. Anthropic’s approach stresses caution, even if it means refusing more questions. OpenAI’s systems aim for broader use, though they accept higher error risk in some hard cases. Both firms said the results were not a ranking, but rather a way to better see trade-offs.
The project also links to wider oversight. Both firms have deals with the U.S. AI Safety Institute for model checks. In addition, OpenAI noted that the results back its launch of GPT-5, which is designed to cut errors and improve safe use.
Using TipRanks’ Comparison Tool, we analyzed several leading companies developing AI chatbots similar to Anthropic’s Claude. This side-by-side view helps investors better understand each stock as well as the broader AI chatbot market.
