Anthropic's Mythos: A Sensitive AI Model Hacked by Third Parties

⚡

Key Takeaways

1The Mythos AI model from Anthropic was accessed by unauthorized users through a subcontractor.

2Mythos, capable of exploiting vulnerabilities in systems, is reserved for companies like Google and Microsoft.

3Anthropic is investigating this unauthorized access, which has not yet shown any impact on its internal systems.

💡Why it matters — This incident highlights the security risks associated with advanced AI models, even when they are strictly controlled.

⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄

Full Analysis

Unauthorized Access to Mythos

The artificial intelligence model Mythos developed by Anthropic, known for its advanced cybersecurity capabilities, has been compromised by a small group of unauthorized users. According to Bloomberg, these users managed to access the model by exploiting a third-party contractor's access and using common online research tools.

Capabilities and Restrictions of Mythos

The Claude Mythos Preview is a versatile model capable of identifying and exploiting vulnerabilities in major operating systems and web browsers. Anthropic has restricted access to this model to a limited number of companies as part of its Project Glasswing, including giants like Nvidia, Google, Amazon Web Services, Apple, and Microsoft. Due to the potential risks of malicious use, Anthropic does not intend to make this model public.

Ongoing Investigation

A spokesperson for Anthropic confirmed to Bloomberg that an investigation is underway regarding this unauthorized access to Mythos via a third-party vendor environment. For now, there is no evidence indicating that this access has affected Anthropic's internal systems.

Incident Details

The unauthorized access reportedly occurred on April 7, coinciding with Anthropic's announcement about making Mythos available to certain companies for testing. The users involved in this unauthorized access are linked to a Discord channel that focuses on unpublished AI models. They used information obtained during a recent data breach at Mercor to locate Mythos online.

Although they accessed Mythos, the group members did not use the model for cybersecurity activities, thus avoiding detection by Anthropic. Other unpublished AI models from Anthropic have also been compromised by this group, according to Bloomberg.

⚡

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.

📰 Voir toutes les actus IA →