Google Accelerates Gemini Model Shipments Amid AI Safety Report Delays

Over two years after being caught off guard by OpenAI's ChatGPT, Google has seriously stepped up its game. In late March, they rolled out Gemini 2.5 Pro, an AI reasoning model that's leading the pack in coding and math benchmarks. And just three months before that, they introduced Gemini 2.0 Flash, which was top-notch at the time. Tulsee Doshi, Google's director and head of product for Gemini, told TechCrunch that this rapid-fire release schedule is all part of their strategy to keep pace with the fast-moving AI world.

"We're still figuring out the best way to release these models and get feedback," Doshi explained. But this faster rollout seems to have a downside. Google hasn't put out safety reports for their latest models, like Gemini 2.5 Pro and Gemini 2.0 Flash, which has folks worried that they're rushing things without being transparent enough.

These days, it's pretty normal for big AI labs, like OpenAI, Anthropic, and Meta, to share safety tests, performance checks, and use cases when they launch a new model. These reports, sometimes called "system cards" or "model cards," were suggested years back by researchers from both industry and academia. In fact, Google was one of the first to propose model cards in a 2019 paper, saying they're key for responsible and transparent machine learning practices.

Doshi mentioned to TechCrunch that they haven't released a model card for Gemini 2.5 Pro yet because they're treating it as an "experimental" release. The idea is to get it out there in a limited way, gather feedback, and improve it before a full launch. She said Google plans to publish Gemini 2.5 Pro's model card when it's widely available, and they've already done some safety testing and adversarial red teaming.

A Google spokesperson followed up, saying safety is still a "top priority" and they'll be releasing more docs about their AI models, including Gemini 2.0 Flash, soon. Even though Gemini 2.0 Flash is out there for everyone, it's still missing its model card. The most recent model card Google released was for Gemini 1.5 Pro over a year ago.

System cards and model cards give out useful info—sometimes stuff that companies might not want to shout about. Like, OpenAI's system card for their o1 model showed it had a habit of "scheming" against humans and secretly chasing its own goals. The AI community generally sees these reports as a way to support independent research and safety checks, but they've become even more important lately. As Transformer pointed out, Google promised the U.S. government in 2023 to release safety reports for all "significant" public AI model launches "within scope." They made similar promises to other governments about being transparent.

There have been efforts at both federal and state levels in the U.S. to set up safety reporting standards for AI developers, but they haven't really taken off. One big attempt was California's SB 1047 bill, which got vetoed and was strongly opposed by the tech industry. There's also been talk of legislation to let the U.S. AI Safety Institute set guidelines for model releases, but now the Institute might face budget cuts under the Trump administration.

It looks like Google's falling behind on some of its promises to report on model testing while pushing out models faster than ever. A lot of experts think this sets a bad example, especially as these models get more advanced and powerful.