Brandon Fletcher
Anthropic is currently investigating potential instances of unauthorized access to its Claude Mythos model, which has gained recognition for its advanced capabilities in identifying cybersecurity vulnerabilities. This revelation was shared with Bloomberg. Reports indicate that a group managed to gain access to this sophisticated model via a third-party contractor portal, utilizing various internet sleuthing tools. However, sources close to the situation suggest that the group’s intentions are benign, focusing on experimentation with the models rather than malicious exploitation.
In an official statement, Anthropic addressed the situation by saying, “We’re investigating a report claiming unauthorized access to Claude Mythos through one of our third-party vendor environments.” This highlights the company’s commitment to maintaining robust security protocols while addressing any potential vulnerabilities associated with its technologies.
The launch of the Claude Mythos Preview occurred earlier this month as part of the highly anticipated “Project Glasswing”, accompanied by significant excitement and media coverage. Anthropic strategically restricted the preview release to a select group of trusted partners, including major tech giants such as Amazon, Microsoft, Apple, and Cisco. Additionally, Mozilla reported that the model contributed to identifying and resolving 271 Firefox vulnerabilities. An increasing number of financial institutions and government entities have expressed interest in accessing this model to enhance their cybersecurity measures and protect their infrastructures.
Nevertheless, multiple unauthorized individuals, reportedly communicating through a private Discord chat, allegedly gained access to Mythos via a developer portal, employing educated guesses to determine the model’s location. This same group might also have gained access to other unreleased models developed by Anthropic, raising concerns about potential security breaches and the dissemination of sensitive information.
The emerging Mythos model has recently drawn attention for its purported capability to detect security flaws within operating systems and internet browsers. This has sparked skepticism among security researchers, alongside growing fears that AI-generated cyber attacks could evolve into a tangible threat. Alex Zenla, CTO of cloud security</b firm Edera, articulated these concerns in a recent interview with Wired. Additionally, it is noteworthy that Anthropic has recently been labeled a “supply chain risk” by the US Department of Defense, although the company has engaged in discussions with the Trump administration to potentially rectify this designation.
Daniel Mercer
