More than two years following Google’s unexpected position on OpenAI’s release of ChatGPT, the company has significantly accelerated its efforts. In late March, Google introduced an AI reasoning model, Gemini 2.5 Pro, which currently leads the industry in several coding and mathematical capability benchmarks. This release followed the debut of another model, Gemini 2.0 Flash, three months earlier, which was considered state-of-the-art at the time.
Tulsee Doshi, Google’s Director and Head of Product for Gemini, explained in an interview with TechCrunch that the increased frequency of model launches is a deliberate effort to keep pace with the fast-evolving AI industry. She commented, “We’re still trying to figure out what the right way to put these models out is — what the right way is to get feedback.”
However, the accelerated release schedule seems to have come with a downside. Google has not yet published safety reports for its latest models, including Gemini 2.5 Pro and Gemini 2.0 Flash, prompting concerns that the company may be prioritizing speed over transparency.
It is now standard for leading AI labs—such as OpenAI, Anthropic, and Meta—to issue reports on safety testing, performance evaluations, and use cases when launching new models. These reports, sometimes referred to as system cards or model cards, were proposed by researchers in both industry and academia several years ago. Google was one of the pioneers to suggest model cards in a 2019 research paper, describing them as a method for responsible, transparent, and accountable practices in machine learning.
Doshi informed TechCrunch that the absence of a model card for Gemini 2.5 Pro is due to its classification as an “experimental” release. The purpose of these experimental releases is to distribute an AI model in a limited manner, obtain feedback, and refine the model before its production launch. Google plans to release the model card for Gemini 2.5 Pro once it is generally available. Doshi also mentioned that safety testing and adversarial analysis have already been conducted.
A Google spokesperson added in a follow-up message to TechCrunch that safety remains a “top priority” for the company, and more documentation regarding its AI models, including Gemini 2.0 Flash, is expected in the future. Currently, Gemini 2.0 Flash, which is available to the general public, lacks a model card. The last model card Google released was for Gemini 1.5 Pro over a year ago.
System cards and model cards can provide valuable information, sometimes revealing critical details that companies do not typically advertise. For instance, OpenAI’s system card for its o1 reasoning model disclosed that it has a tendency to act against human interests.
Overall, the AI community views these reports as sincere attempts to support independent research and safety evaluations. However, their importance has grown in recent years. As previously noted by Transformer, Google pledged to the U.S. government in 2023 that it would publish safety reports for all “significant,” public AI model releases “within scope.” Similar commitments were made to other governments, promising to maintain “public transparency.”
There have been legislative efforts at both federal and state levels in the U.S. to establish safety reporting standards for AI model developers, although these efforts have seen limited success. Notably, a California bill, SB 1047, was vetoed despite intense opposition from the tech industry. There is legislation proposed to empower the U.S. AI Safety Institute to set guidelines for model releases, although the institute is currently facing potential budget cuts under the Trump Administration.
It appears that Google is lagging on some commitments to report on model testing while simultaneously increasing the speed of model releases. Many experts argue that this sets a concerning precedent, particularly as models grow more capable and sophisticated.
