Google has announced a new artificial intelligence model, Gemini 4 Argon, and called it its most powerful yet. It writes code, does research, writes text, analyses long videos and charts. But what it boasts about most is one thing: cybersecurity. According to the company, Argon can "autonomously find, verify and patch critical vulnerabilities in software".

Ordinary users won't see it for now. The model is going only to a select group of security partners through Fairwind, Google's security programme, and has been trained specifically for defensive work. Inside the company it is already used daily - employees use it to hunt bugs in code and migrate entire codebases to new systems.

Next comes the obligatory part of every announcement like this: the table. Google claims Argon "significantly" outperformed OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models, citing Vals, an AI model evaluation startup, on whose index Argon currently ranks first. OpenAI recently presented Astra as its best model yet. Earlier this year Anthropic presented Fable in almost the same words. Every new model is "the best yet" - until the next one comes out.

The irony is familiar. The same labs racing to release ever more powerful models are the loudest in warning that artificial intelligence could get out of control. And a tool that finds software flaws by itself is just as interesting to whoever patches them as to whoever would like to break in. Is that why Argon is staying behind the door of selected partners for now?

The market picture is clearer than the tables. Google was long seen as the company "lagging behind" in the race, and in August it announced that the Gemini app has more than a billion monthly users. That put it level with OpenAI, since ChatGPT recently hit the same number. Who will hold the keys to the software everything else runs on - that question doesn't show up in any table.