In a recent development, Anthropic has opted to restrict access to its advanced AI model, Claude Mythos, which is specifically designed to identify security flaws in software systems. This decision comes on the heels of a leak regarding Anthropic's own content management system, prompting the company to reconsider the public release of Mythos.
On Tuesday, Anthropic unveiled Claude Mythos Preview, an unreleased AI model that holds the promise to significantly reshape the landscape of cybersecurity. According to the company, Mythos has already detected thousands of high-severity vulnerabilities across major operating systems and web browsers. While these findings could be beneficial for enhancing security, Anthropic has expressed serious concerns that malicious actors could exploit the technology for harmful purposes, potentially leading to severe repercussions for economies, public safety, and national security.
To mitigate these risks, Anthropic has decided to limit access to Claude Mythos Preview to a select group of organizations. Among those granted access are notable tech giants such as Amazon Web Services, Apple, Google, Microsoft, and NVIDIA, as well as cybersecurity firms like CrowdStrike and Palo Alto Networks. This initiative, referred to as Project Glasswing, aims to bolster critical software infrastructure through the responsible use of AI. Additionally, over 40 other organizations that focus on maintaining critical software infrastructure will also have access to the model.
Dianne Penn, Anthropic’s head of research product management, emphasized in an interview with CNBC that the decision to restrict access followed extensive internal discussions. The goal is to provide cyber defenders with a head start in identifying and addressing vulnerabilities before they can be exploited. Anthropic's CEO, Dario Amodei, echoed these sentiments in a post on social media, stating, "Rather than release Mythos Preview to general availability, we’re giving defenders early controlled access in order to find and patch vulnerabilities before Mythos-class models proliferate across the ecosystem."
Amodei further cautioned about the potential dangers of mishandling this powerful technology, noting, "The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities." While Claude Mythos Preview excels at pinpointing cybersecurity flaws, it is primarily a general-purpose model. Currently, Anthropic has no plans for a broad release; instead, the company is focused on determining how to safely deploy Mythos-class models to a wider audience.
This initiative reflects a growing recognition within the tech community of the dual-edged nature of advanced AI. While such technology can enhance security measures, it also poses significant risks if misused. As the industry grapples with these challenges, Anthropic's cautious approach serves as a reminder of the responsibilities that come with developing powerful AI tools.
Source: Mashable News