Experts warn of safety gaps in Google Gemini 2.5 Pro technical report
Experts have criticized Google's technical report on its Gemini 2.5 Pro AI model for lacking sufficient safety assessment details, making it difficult to verify the company's safety commitments and evaluate potential risks from the model.
What happened
Google published a technical report with results of internal safety assessments of its Gemini 2.5 Pro model weeks after releasing the model publicly.
Experts stated the report lacks important details, including missing references to Google's safety framework from the previous year designed to identify AI capabilities that could cause severe harm.
Peter Willford, co-founder of the Center for AI Policy, said the report is scattered with minimal information and was released after weeks of public availability, making it impossible to verify if Google fulfills its public commitments or to assess the safety of its models.
Reported impact
Thomas Woodside, co-founder of the Source AI Project, expressed concern that the report does not demonstrate Google's commitment to providing timely additional safety assessments, noting the company had not published dangerous capability test results since June of the previous year.
Index analysis
The AI community views technical safety reports as genuine efforts to support independent research and safety evaluations, but multiple experts expressed disappointment with the lack of substantial details in the Gemini 2.5 Pro report.
Governance controls
Google published a technical report and developed a safety framework the previous year to identify AI capabilities that could cause severe harm.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.