The American giant Google unveiled Gemini 4 on Wednesday, its new cutting-edge artificial intelligence (AI) model, but for the moment reserves access to a restricted circle of cybersecurity experts, via a program called Fairwind.
Google is thus following in the footsteps of its rivals Anthropic and OpenAI, which also began this year to reserve their latest models for a limited number of trusted partners, due to their ability to identify computer vulnerabilities likely to be exploited by malicious actors.
“There is a lot of talk about our next model (!), so I wanted to give an overview as soon as possible,” wrote on
The group promises to quickly open this model, called Gemini 4 Argon, to businesses and the general public, starting with its paying customers, without giving a date.
“Argon has safeguards adapted to cutting-edge models and we deploy them responsibly,” assured Sundar Pichai. “We are going to make it available as quickly and as carefully as possible. So be patient.”
In April, Anthropic reserved Claude Mythos for around fifty organizations responsible for securing critical software, including Google. At the beginning of September, OpenAI limited access to GPT-6 Astra to approved companies, its first model classified as a critical cybersecurity risk according to its own criteria.
Since the spring, the performance of these cutting-edge models, capable according to their designers of uncovering serious IT flaws on their own, or even exploiting them, has shaken institutions, worried about possible attacks against critical infrastructures (electricity networks, banks, administrations) if such tools were made public before the breaches are plugged.
Fears revived by series of incidents
A series of incidents since the summer has reignited fears about the dangers of AI. In July, hundreds of OpenAI AI agents, software capable of acting autonomously, penetrated the servers of the Hugging Face platform. Anthropic, Meta and Google have reported similar incidents, which have renewed calls to slow the development of the technology.
Tuesday at the White House, the bosses of the sector, including Sundar Pichai, adopted a voluntary code of good conduct, “morally binding” according to Donald Trump, to respond to these risks.
Washington fights any national or international regulation, in order to maintain its competitiveness against China. A presidential decree in June, however, created a voluntary federal review of the most advanced models and Google said Tuesday that it had submitted Argon to this procedure.
