The world of cybersecurity is in turmoil: Anthropic's latest AI model, Mythos, has discovered serial computer vulnerabilities... which have sometimes existed for more than 20 years. For the moment, only a few “trusted” actors have access to it. But after?

Published on: 04/15/2026 - 6:09 p.m.

Warnings are multiplying about Mythos, Anthropic's cybersecurity "expert" AI. This AI model, capable of testing the security of computer systems and finding flaws, is worrying all the way to the top of states. On Monday April 13, the British government called on financial institutions to urgently examine the capabilities of this algorithm and to strengthen their IT defenses.

David Solomon, CEO of the American banking giant Goldman Sachs, assured Tuesday that he was “hyper vigilant and aware” of the risks posed by this new tool. A few days earlier, it was the American administration which brought together several leaders of banks and institutions such as the Federal Reserve to discuss the challenges posed by Mythos and other new AI models that have become masters of cybersecurity.

A “mind-blowing” number of vulnerabilities discovered

However, few people have yet been able to see what Mythos is capable of.

In parallel with the official announcement of the launch of this AI on April 7, Anthropic decided to set up a small committee – called Project Glasswing – bringing together only around fifty organizations.

Tech giants, such as Amazon or Google, big names in cybersecurity, like CrowdStrike or Palo Alto Networks, and free software players, like the Linux Foundation, are currently the only ones who can come up against the Anthropic creature.

For its creator, Project Glasswing is necessary, because such a powerful tool should not fall into the wrong hands first. An approach which was received with caution. Is Mythos mytho? And is Anthropic cultivating secrecy to create buzz around its latest product?

To display this YouTube content, it is necessary to authorize audience measurement and advertising cookies.

An extension in your browser appears to be blocking the loading of the video player. To be able to watch this content, you must disable or uninstall it.

“I would have preferred that it was just marketing and that Mythos was less powerful than advertised. It would have really made our lives easier these days,” regrets Sylvestre Ledru, director of engineering for Mozilla. The foundation that manages the Firefox browser can have access to Mythos to test its products and, "with Firefox, one of the most attacked and secure tools in decades, we thought we had seen and heard everything when it came to security flaws. But the number of vulnerabilities discovered by Mythos - and not only among us - is truly mind-blowing," he admits.

This is what Sylvestre Ledru calls “the Mythos revolution”. “Its ability to quickly, autonomously and reliably discover vulnerabilities across a broad spectrum of software and network infrastructure is the game changer,” says Lukasz Olejnik, a technology security consultant and research associate at King’s College London.

27-year-old computer flaws

“Mythos represents an important evolution compared to more targeted models which require very precise instructions and constant human support to discover security vulnerabilities,” adds Sven Herpig, specialist in cybersecurity issues and emerging risks at Interface, a German NGO studying the societal impact of new technologies.

The scale of Mythos' discoveries, detailed by Anthropic, is enough to make one dizzy. This AI thus identified a flaw that had gone unnoticed for 27 years in an operating system used by millions of people. “Certain flaws that he found existed for more than a decade,” Sylvestre Ledru also recognizes.

Also read: ChatGPT artificial intelligence and the democratization of cybercrime

These are not obscure security vulnerabilities that pose no real IT security problem. “Certain vulnerabilities can be qualified as critical,” assures the Mozilla expert. In other words, a malicious actor getting their hands on this vulnerability vault could do damage.

For the experts interviewed, this is the whole point of Anthropic’s cautious approach. “For the moment, it is the defenders who have the advantage because they are the only ones with access to it, but as soon as Mythos is made public, the attackers risk gaining the upper hand,” warns Sven Herpig.

Countdown to a “new era” of cybersecurity

For this expert, one of the main risks comes from attackers sent to the cyber front by States, whether hackers working for Russia or China, or even for Western countries. What will happen if Mythos discovers flaws in other states' critical infrastructure for them?

But the danger also comes from the bottom of the cybercriminals' food chain. “Mythos is able to find a vulnerability and then create a method to exploit it without the need for human intervention. This will lower the entry ticket for hackers, who will no longer need to have very high technical knowledge to carry out sophisticated attacks,” assures Jason Nurse, cybersecurity expert at the University of Kent.

The establishment of Project Glasswing thus constitutes the start of a countdown before “the advent of a new reality of cybersecurity”, affirms Lukasz Olejnik. All the participants in this committee are engaged in a race to correct all the vulnerabilities discovered "before Anthropic decides to make Mythos public or before another player in the sector develops its own solution", underlines Sven Herpig.

OpenAI, the creator of ChatGPT, announced on Tuesday the launch of GPT-4.5 Cyber, its rival to Mythos. Here too, the availability of this AI is restricted to a few selected organizations. The experts interviewed also wonder if the Chinese are preparing to launch their solution and if it will be deployed with the same caution.

“What about all the software developers who are not associated with these small groups who have the right to test Mythos?” warns Jason Nurse. Not to mention that "certain parts of the Internet rely on free software which is maintained by only one or two volunteers. It is impossible for them to correct all the vulnerabilities that Mythos could find", underlines Sven Herpig.

It is therefore a safe bet that this interlude during which only those whom Anthropic describes as "good actors" can work with Mythos is not enough. Ultimately, "there will probably be a deluge of patches to be installed very quickly by companies and organizations around the world," says Lukasz Olejnik.

Mythos can thus be considered as the proverbial Pandora's box from which a thousand and one new cyber dangers risk emerging. But there is also the more optimistic scenario in which Mythos gives a big kick to the anthill, forcing everyone to strengthen "software security in general", concludes Sylvestre Ledru.

Keywords associated with the article

The content you are trying to access does not exist or is no longer available.