Google Criticized for Skipping Safety Reviews on Gemini 2.5 Pro

Google is drawing widespread criticism from researchers and AI experts over the release of its technical safety report for its latest artificial intelligence model, Gemini 2.5 Pro. The report, which was made public shortly after the model’s launch, confirms that Google performed some internal safety checks. However, many in the AI community argue that the report lacks sufficient detail, raising serious questions about the company’s transparency and overall commitment to Gemini AI safety.

Unlike its competitors, Google has a policy of releasing technical safety reports only after a model is no longer considered “experimental.” More importantly, these reports omit critical evaluations known as “dangerous capability” tests, which assess a model’s potential for misuse or harmful behavior. These findings are reportedly reserved for internal audits and are not made publicly available, contributing to growing unease within the industry.

Peter Wildeford, co-founder of the Institute for AI Policy and Strategy, described the safety report as “very sparse” and lacking the depth necessary to assess the model’s real-world impact. He also criticized the timing of the release, noting that it came weeks after Gemini 2.5 Pro was launched, making it difficult to verify whether safety protocols were followed during deployment.

This situation has intensified the Google AI controversy, with many experts calling for greater openness and stricter oversight. The Gemini 2.5 Pro criticism is not just about the lack of technical data but also about Google skipping safety protocols that are becoming standard practice among leading AI developers.

As the AI landscape becomes more complex and powerful, the call for robust safety evaluations and public transparency is louder than ever. Google's handling of Gemini 2.5 Pro may well shape how future AI governance standards are formed.