An AI too powerful to be public?

Have you seen? Anthropic has just made an unprecedented decision: not to publicly release its latest model, Claude Mythos Preview. Not for lack of...

An AI too powerful to be public?

An AI too powerful to be public? 1024 683 Jean-Marc Segati

Have you seen? Anthropic has just made an unprecedented decision: not to publicly release its latest model, Claude Mythos Preview.

Not for lack of trust. Because of excessive lucidity!

In a few weeks of testing, Mythos identified thousands of vulnerabilities in the major operating systems and browsers — some of which had existed for more than 25 years, invisible to decades of human audits and automated testing.

The problem isn’t that it’s good. It’s that it’s too good to be freely available.

“Too powerful,” what does that mean concretely?

It’s not a question of creativity or productivity. It’s the moment when a model allows anyone — including poorly qualified malicious actors — to cause disproportionate damage: industrializing cyberattacks, scanning millions of lines of code, generating functional exploits. Automatically. Without supervision.

The real warning signal: the control switch. When the tool moves from “powerful assistance under human supervision” to “uncontrollable autonomous capability,” the question isn’t about performance anymore. It’s: who decides, in what framework, with what guardrails?

What Anthropic has chosen to do

Rather than a public launch, access is limited to about fifty organizations (Microsoft, CrowdStrike, Apple, Palo Alto Networks…) via Project Glasswing. Mission: use this power to strengthen global cybersecurity, not to weaken it.

This is a major precedent. The first time since GPT-2 in 2019 that a leading AI lab has voluntarily withheld a model.

What this changes for us

→ The “same AI for everyone” narrative is over. Some models will never be public. Access to advanced capabilities will become a differentiating competitive advantage.

→ The limit that matters isn’t brute power. It’s human control over the machine. An AI without a use framework, without governance, without a human decision-maker in the loop, is a weapon — whatever the initial intention.

→ For businesses, the question is no longer “which AI to adopt?” but “what governance framework should be put in place to use these capabilities responsibly?”

The most powerful AI in the world is useless if no one is at the wheel.

#AI #ArtificialIntelligence #Cybersecurity #Anthropic #ClaudeMythos #AIGovernance #ResponsibleAI #Leadership #Innovation #Tech #FutureOfAI #CyberSecurity #ProjectGlasswing

)

Préférences

Quand vous visitez un site, certaines informations sont gardées en mémoire par votre navigateur sous forme de cookies. Vous pouvez changer ici vos préférences. Il se pourrait que bloquer ces cookies ait un impact sur votre expérience utilisateur et sur les services que nous offrons.

Cliquez pour activer/désactiver google analytics
Cliquez pour activer/désactiver google font
Cliquez pour activer/désactiver google map
Notre site web utilise des cookies pour rendre son site Web fonctionnel avec l'intention de rendre votre visite la plus agréable possible. Les cookies sont inoffensifs. En continuant à surfer sur notre site web, vous acceptez l'utilisation de cookies. Définissez vos préférences ou acceptez leur utilisation.