Google has been criticized for opacity in the report on Gemini 2.5 Pro

< img src = "/uploads/blogs/ad/c6/ib-ib-free4c_1a2a8067.jpg" Alt = "Google was criticized for opacity in the report on Gemini 2.5 Pro"/> ~ ~ ~ ~ < P class = "" Data-Start = "133" Data-end = "468" > Google was under the fire of criticism due to insufficient openness and lack of details in the technical report dedicated to its flagship shi model GEMINI 2.5 PRO. Experts believe that the published document does not contain sufficient information for OB & Rsquo; a centered analysis < P class = "" Data-Start = "470" Data-end = "786" > The report was published after the model was launched, which caused dissatisfaction with researchers and public organizations. According to Peter Wildeford, co -founder of the Institute of Policy and Strategy in the field of AI, this practice makes it impossible to independently check Google promises and evaluate the reliability of its products.

~ ~ < P class = "" Data-Start = "788" Data-End = "1125" >Particular concern was caused by the lack of information about the internal risk verification system & mdash; Frontier Safety Framework (FSF). Google has presented this tool last year to identify the dangerous AI abilities, but the results of the relevant tests were not included in the report, but are stored separately and are usually not published.

< P class = "" Data-Start = "1127" Data-End = "1483" > Analysts note that Google, which has previously advocated transparent reporting standards in the field, is now deviating from its own principles. The situation is also complicated by the fact that the company has not yet provided a report on the recently presented model Gemini 2.5 Flash. Google assured that the document with & Rsquo; it will be & laquo; in the near future & raquo; < P class = "" Data-Start = "1485" Data-end = "1684" > However experts indicate that this situation & mdash; Not single. For example, Meta also received criticism for the floor report on LLAMA 4, and Openai did not publish the GPT-4.1 risk analysis at all. ~ < P class = "" Data-Start = "1686" Data-End = "1997" > & laquo; we observe a disturbing tendency to reduce standards in reporting, & mdash; said Kevin Bankston, senior adviser to the Center for Democracy and Technology. & mdash; And when the time to test new models is reduced from months to a few days, according to some companies, this approach becomes especially dangerous & raquo;.