Anthropic Restricts Access to Powerful ‘Claude Mythos’ AI Amid Cybersecurity Concerns
Anthropic has unveiled Claude Mythos Preview, a highly advanced artificial intelligence model demonstrating unprecedented capabilities in identifying and exploiting software vulnerabilities. Due to the potential risks associated with the model’s power, the company is limiting public access and instead launching a targeted cybersecurity initiative known as Project Glasswing.
Announced on April 7, 2026, Claude Mythos Preview is described as a general-purpose language model that is “strikingly capable” at computer security tasks. According to technical details released by the company, the model can identify and exploit zero-day vulnerabilities in open-source codebases and reverse-engineer exploits within closed-source software. It has also proven capable of turning N-day vulnerabilities—known flaws that have not yet been widely patched—into functional exploits.
The model employs sophisticated exploitation techniques, including ROP attacks and JIT heap sprays. While researchers note these primitives are well-understood, the novel ways Mythos Preview identifies vulnerabilities and chains them together represent what Anthropic calls a “watershed moment for security.”
To mitigate the risk of the technology falling into the hands of malicious actors, Anthropic is restricting the model’s rollout. Under Project Glasswing, a select group of approximately 50 companies will be granted access to the model for defensive security purposes. Initial launch partners include industry giants such as Microsoft, Amazon Web Services, Apple, Google and Nvidia, as well as cybersecurity firms CrowdStrike and Palo Alto Networks.
The decision to restrict the model follows a period of internal deliberation. “We really do view this as a first step for giving a lot of cyber defenders a head start on a topic that will be increasingly important,” Dianne Penn, Anthropic’s head of research product management, stated in an interview on April 7, 2026.
The announcement follows a period of market volatility. Descriptions of the model’s advanced capabilities were discovered by Fortune in a publicly accessible data cache in late March 2026. This leak triggered a sell-off in cybersecurity stocks as investors reacted to the potential risks posed by an AI capable of automating complex cyberattacks.
The restricted rollout underscores the growing tension between AI innovation and global security. By partnering with critical infrastructure and software providers, Anthropic aims to assist secure the world’s most critical software and prepare the industry for the defensive practices required to counter AI-driven threats. This strategic move highlights the increasing economic and security relevance of AI-powered vulnerability research.
The development of Claude Mythos has sparked significant debate regarding the safety of high-capability models, leading some to question how dangerous the model actually is and why authorities in the U.S. Are concerned. The situation has been described as a case where Anthropic created an AI so potent that it decided not to release it to the general public.
As Wall Street remains on alert regarding the potential for AI to reveal thousands of cyber flaws, the industry is closely watching how Project Glasswing will influence the balance of power between cyber attackers and defenders.
For those seeking to understand what Mythos is and why it concerns U.S. Authorities, the focus remains on whether restrictive access is a sufficient safeguard against the automation of high-level cyber warfare.
