Experts have criticized Google for the lack of transparency and minimal information in the technical report on Gemini 2.5 Pro. They believe that the document, without full data, does not give an adequate idea of the possible risks of the new AI model.
Image source: Google
The report comes just weeks after the launch of Gemini 2.5 Pro, Google’s most powerful model to date. While such documents are typically considered an important part of ensuring artificial intelligence (AI) security and help independent researchers conduct their own risk assessments, the report in this case was “very thin,” TechCrunch writes.
«The document is minimal in information and was released after the model was publicly available, said Peter Wildeford, co-founder of the Institute for AI Policy and Strategy. “There’s no way to verify that Google is delivering on its public promises, which means there’s no way to assess the safety and reliability of its models.”
A separate criticism was the lack of mention of Google’s internal Frontier Safety Framework (FSF), which it introduced last year to identify potentially dangerous AI capabilities. The new report does not include the results of tests on the model’s dangerous capabilities — data that Google stores separately and does not typically publish with the main document.
Experts believe that Google, which once advocated for standardized AI reporting, is now itself moving away from its principles. Particularly concerning is the fact that the company has yet to release a report on its recently announced Gemini 2.5 Flash model. However, a Google spokesperson told TechCrunch that the document will be released soon.
It is noted that Google’s situation is apparently part of a larger trend. For example, Meta
has also recently come under fire for its superficial risk analysis of its latest AI model, Llama 4, and OpenAI has not provided a report on its GPT-4.1 line at all.
«“We’re clearly seeing a race to the bottom,” said Kevin Bankston, senior adviser for AI governance at the Center for Democracy and Technology. “And given reports that other companies, including OpenAI, are cutting testing time from months to days, this level of reporting from Google is alarming.”