Experts Say Google’s Gemini 2.5 AI Report Falls Short on Key Safety Details

Google Faces Criticism Over Sparse AI Safety Reporting for Gemini 2.5 Pro

Weeks after launching its most advanced AI model yet, Gemini 2.5 Pro, Google released a long-awaited technical report outlining internal safety evaluations. But according to AI safety experts, the report lacks key details, casting doubt on Google's commitment to transparency and raising broader concerns about the industry's race to release increasingly powerful models with minimal public oversight.

A Lacking Safety Report

While Google’s publication of a technical report is, on the surface, a welcome gesture toward openness, experts say the Gemini 2.5 Pro report is thin on substance. Crucially, it omits information related to Google's Frontier Safety Framework (FSF)—an initiative introduced last year to identify and mitigate severe risks associated with advanced AI systems.

“This [report] is very sparse, contains minimal information, and came out weeks after the model was already made available to the public,” said Peter Wildeford, co-founder of the Institute for AI Policy and Strategy, in comments to TechCrunch. “It’s impossible to verify if Google is living up to its public commitments.”

Wildeford and other researchers have voiced concern that the report does not clarify how the company is evaluating potentially dangerous capabilities, nor whether those evaluations are happening in a timely and structured manner.

Delays and Gaps in Safety Transparency

Thomas Woodside, co-founder of the Secure AI Project, welcomed the publication but called out the inconsistency and delay in Google’s safety disclosures. Notably, the company hasn't released results from its dangerous capabilities testing since June 2024—a report that covered a model released four months earlier.

Even more concerning, no report has yet been issued for Gemini 2.5 Flash, a smaller and more efficient variant released just last week. A Google spokesperson has said the Flash report is “coming soon,” but provided no timeline.

Woodside emphasized the need for frequent and proactive updates, especially for models that haven’t yet been publicly deployed but may still present serious risks. “Those updates should include the results of evaluations for models that haven’t been publicly deployed yet,” he said.

Broader Trend: A Transparency “Race to the Bottom”

The criticism aimed at Google comes amid a wider industry trend of reduced transparency among major AI developers. Meta’s recent safety evaluation for Llama 4 was similarly light on technical and risk-related content, and OpenAI has yet to publish a safety report for its GPT-4.1 series.

“This meager documentation for Google’s top AI model tells a troubling story of a race to the bottom on AI safety and transparency,” said Kevin Bankston, senior adviser on AI governance at the Center for Democracy and Technology. He also referenced reports that some AI labs have cut their safety testing timelines dramatically—from months to just days—in order to speed up deployment.

These developments are particularly troubling given Google’s prior commitments. Two years ago, the company pledged to the U.S. government and international regulators that it would publish safety reports for all “significant” AI models and provide “public transparency” around their capabilities and risks.

Google Defends Internal Processes

Google, for its part, maintains that it conducts rigorous internal safety testing, including adversarial red teaming, before launching any model. However, these processes are largely undocumented in public-facing materials, making it difficult for outside researchers, policymakers, or users to assess the robustness of its safety protocols.

The company’s practice of only publishing safety reports after models are no longer considered experimental also limits timely transparency. Critics argue this approach delays public scrutiny until well after a model is already in widespread use.

What’s at Stake

As AI capabilities scale rapidly, trust in the institutions that develop and deploy them is critical. Sparse safety reports, shifting definitions of transparency, and missed deadlines on key disclosures raise pressing questions about how effectively the industry is policing itself—and whether regulatory frameworks are keeping up.

Experts argue that meaningful transparency isn’t just about releasing a report—it’s about demonstrating accountability, clarifying risks, and informing both the public and policymakers. Without that, trust in AI’s societal integration could erode just as fast as its capabilities expand.

Stay tuned to The Horizons Times for expert analysis on AI safety, industry standards, and how leading tech firms are shaping the future of artificial intelligence accountability.