## Google’s Gemini 2.5 Pro Safety Report Under Fire for Lack of Transparency
Google’s release of its Gemini 2.5 Pro technical report, weeks after the AI model’s launch, has sparked criticism from experts who claim it lacks crucial safety details. The report, intended to showcase internal safety evaluations, is viewed as too sparse, making it difficult to assess the potential risks associated with the model.
Technical reports are typically seen as a valuable resource, offering insights into AI models that companies might not otherwise publicize. These reports support independent research and safety assessments, fostering transparency within the AI community.
Google’s approach differs from some competitors, as it only releases these reports after a model is deemed to have moved beyond the “experimental” stage. Furthermore, the company doesn’t include findings from all “dangerous capability” evaluations in these reports, reserving them for separate audits.
However, several experts speaking with TechCrunch expressed disappointment with the Gemini 2.5 Pro report’s limited scope. Notably, the report fails to mention Google’s own Frontier Safety Framework (FSF), introduced last year to identify future AI capabilities with the potential for “severe harm.”
Peter Wildeford, co-founder of the Institute for AI Policy and Strategy, stated that the report contains “minimal information” and was released “weeks after the model was already made available to the public.” He argues that this lack of transparency makes it “impossible to verify if Google is living up to its public commitments and thus impossible to assess the safety and security of their models.”
Thomas Woodside, co-founder of the Secure AI Project, acknowledges the report’s existence but questions Google’s commitment to timely safety evaluations. He points out that the last time Google released results of dangerous capability tests was in June 2024, for a model announced months prior.
Adding to the concerns, Google has yet to release a report for Gemini 2.5 Flash, a more efficient model announced just last week. A spokesperson has stated that a report for Flash is “coming soon.”
Woodside hopes that this promise signals a shift towards “more frequent updates,” including “the results of evaluations for models that haven’t been publicly deployed yet.” He stresses that these pre-deployment evaluations are crucial since those models could also pose serious risks.
Google isn’t alone in facing scrutiny. Meta’s Llama 4 safety evaluation has also been criticized for being “skimpy,” and OpenAI opted not to publish a report for its GPT-4.1 series.
These instances raise concerns, particularly given Google’s past assurances to regulators to maintain high standards of AI safety testing and reporting. Two years ago, Google pledged to the U.S. government to publish safety reports for all “significant” public AI models “within scope,” and later made similar commitments to other countries.
Kevin Bankston, a senior adviser on AI governance at the Center for Democracy and Technology, describes the trend of sporadic and vague reports as a “race to the bottom” on AI safety. He notes that this, combined with reports of shortened safety testing times at competing labs, paints a “troubling story of a race to the bottom on AI safety and transparency as companies rush their models to market.”
Google has stated that, while not detailed in its technical reports, it does conduct safety testing and “adversarial red teaming” for models before release. However, the lack of comprehensive public documentation continues to fuel concerns about the responsible development and deployment of increasingly powerful AI systems.