Updates·September 30, 2026, 22:00

Google holds back new AI model over fears of misuse for hacking

AI-generated and checked against the sources listed below.

Google has launched a powerful new AI model called Gemini 4 Argon that can find and fix security holes in software on its own. However, the model is only rolling out gradually, because the same abilities could in principle be misused for hacking attacks.

AI-generated image

Google has released a new version of its AI model Gemini, called Gemini 4 Argon. The model is especially good at finding, confirming and fixing security holes in computer programs, a task that normally requires specialized experts.

Instead of making the model available to everyone right away, Google is rolling it out in phases.

Only a select few get access first

The first to get access are trusted cybersecurity experts through what Google calls the Fairwind program. Here, government agencies, operators of critical infrastructure and large technology companies, among others, can join, provided they follow strict rules for its use.

Later, Argon will open to paying Google customers and Google AI Ultra subscribers. Regular developers, businesses and consumers will have to wait longer before they can use the model; no fixed date has been set for that.

The same abilities can be used for attacks

The reason for the cautious rollout is that the abilities that make Argon good at detecting and closing security holes could in principle also be used by hackers to find weaknesses and attack systems, for example at hospitals, banks or government agencies.

Google has therefore built several layers of security into the model. It is monitored for signs of misuse, and the system can intervene and stop actions the user has not asked for. There are also built-in defenses against so-called prompt injection, where someone tries to trick an AI model into carrying out hidden commands.

In tests, Argon performs on par with or better than competing models such as GPT-6 Astra from OpenAI on several benchmarks that measure the ability to find and fix bugs in code, and according to Google at a lower operating cost.

For ordinary Danes, the news mainly means that AI tools are now getting better at detecting security holes before criminals do, but that the companies behind the technology are at the same time worried that the same tools could be turned against us.

What it means for you

Right now, it is primarily IT security professionals and large companies that benefit from Argon. If you work in operations, development or security, it may make sense to keep an eye on when your company can apply for access via Google or the Fairwind program. As a regular user, you can't do anything concrete yet, but you can expect the software you use to be better tested for security holes in the future. As always, stick to good habits like updates and strong passwords while the technology rolls out.

Sources

More on this topic

Get the week's AI news in your inbox

Choose your level, topics and length. One email a week, unsubscribe at any time.

Subscribe to Promptly Newsletter
PromptlyNewsletterRSSLog in

The news on aijour is AI-generated and checked against the cited sources.