On March 26, cyber sleuths found a large trove of unpublished data of Anthropic AI available publicly on the company’s content management system (CMS). Soon enough, Anthropic retrieved the unpublished files but not before the world had a sneak peek at a folder named ‘Claude Mythos’.

This was the first time the world heard of the term Claude Mythos—apparently the most advanced Artificial Intelligence (AI) model developed by Anthropic that has now rung alarm bells in Washington D.C. over cybersecurity risks.

Also Read: Claude Mythos Preview: US Officials Warn Banks Over Cybersecurity Risks

Post the leak, Anthropic claimed that Claude Mythos, their frontier AI model has become so effective at finding software flaws that the company is limiting access to a small group of technology partners and government-facing organizations.

But because most of the company’s claims remain unverified, the emergence of “Claude Mythos” has triggered two parallel debates: whether frontier AI is becoming a genuine cyber-security threat, and whether AI companies may be overstating those dangers to build mystique around their latest systems.

In this article, we examine Claude Mythos AI model, Anthropic’s claims of LLM’s risks, their Project Glasswing and their paradoxical marketing pitch of Claude Mythos.

What has Anthropic Claimed About Claude Mythos?

Anthropic has claimed that Claude Mythos has demonstrated “a leap in coding skills” and found vulnerabilities in software systems including cases that survived decades of human review and millions of automated security tests. However, there has not been any independent verification of the claims made by Anthropic related to Claude Mythos.

The AI startup has also claimed that it is participating in ongoing discussions with U.S. government officials about Claude Mythos Preview and its offensive and defensive cyber capabilities.

“Governments have an essential role to play in helping maintain AI lead and in assessing and mitigating the national security risks associated with AI models. We are ready to work with local, state and federal representatives to assist in these tasks,” read a statement from Anthropic on Claude Mythos Preview announcement.

The statement comes in the wake of the U.S. Department of Defense (Pentagon) officially designated Anthropic as a “supply chain risk” in March 2026 after its CEO Dario Amodei refused to let the company’s AI model be used for domestic mass surveillance and advanced military operations.

What is Claude Mythos Preview?

Claude Mythos Preview is Anthropic’s claimed “most advanced frontier model” to date, as per the company’s claims. Apart from exhibiting capabilities substantially beyond those of any previously trained model in areas such as software engineering, reasoning, computer use, knowledge work, and research, Mythos Preview has also demonstrated particularly exceptional cybersecurity capabilities.

AI FrontPage has reached out to Anthropic for a detailed comment on the claimed capabilities of Claude Mythos and its independent verification. The article will be updated once Anthropic replies to us.

As per Anthropic, Claude Mythos Preview has identified thousands of zero-day vulnerabilities—flaws that were previously unknown to developers. Vulnerabilities were found in every major operating system and web browser, many of them critical.

Among the most notable were a 27-year-old vulnerability in OpenBSD—widely regarded as one of the most secure operating systems—and a 16-year-old vulnerability in FFmpeg, a widely used video encoding and decoding library. Mythos Preview also autonomously discovered and chained together multiple vulnerabilities in the Linux kernel. These vulnerabilities were responsibly disclosed and have since been patched, said Anthropic.

This large increase in capability has led Anthropic to decide against making the model generally available. Instead, it is being deployed in a controlled manner as part of a defensive cybersecurity initiative. It is precisely this level of capability that has prompted Anthropic to rethink how such systems should be deployed.

What is Project Glasswing?

Two weeks after the first leak, Anthropic announced Project Glasswing– a collaboration with industry titans like Amazon Web Services, Apple, Google, NVIDIA, CrowdStrike, Broadcom among others.

Under Project Glasswing, Anthropic will provide “Claude Mythos Preview”–a limited edition of its frontier LLM model to be used by the project partners, “to find and fix vulnerabilities or weaknesses in their foundational systems.”

Anthropic has offered a commitment of $100M in “model usage credits” to the partners and intends to charge $25/$125 per million input/output tokens afterwards.  However, there has not been any public acknowledgement of Anthropic’s Project Glasswing collaboration by its supposed partners including rivals Google and Apple.

Why are Experts Skeptic About Claude Mythos?

According to a report published in Fortune on March 26, when cyber sleuths discovered unpublished data of Anthropic on its CMS system, the files also had information about an upcoming, invite-only retreat for the CEOs of European companies to be held later this year in U.K.

Yann LeCun, former Chief Scientist at Meta and co-founder of Advanced Machine Intelligence (AMI), has been one of the harshest critics of Anthropic’s statements on Claude Mythos, and has compared it to “self-delusion.” He was replying to a tweet by Clement Delangue, CEO of Hugging Face, who also tried to downplay Anthropic’s claimed capabilities.

The fact that Anthropic has decided not to release Claude Mythos for general public because it is apparently “too dangerous” and make it available only for companies, show a kind of “fear profiteering” campaign that critics are calling out

Conclusion: AI’s role in Cybersecurity

Today all industries make use of and run on software. However, flaws and weaknesses in software can go unnoticed for years but latest frontier AI models, have drastically reduced the cost, effort, and knowledge required to find and exploit such vulnerabilities.

The possibilities are many, it can increase the frequency of cyberattacks, leading to serious economic loss, exposure of sensitive data, and risks to vital systems such as corporate networks, healthcare systems, energy infrastructure, and government systems.

It is in this context that AI’s role in cybersecurity becomes particularly significant. AI models and how they can be used for both defensive and offensive purposes is one of the most critically important aspects of cybersecurity today.

Anthropic sees Project Glasswing as the beginning of a long-term effort where partners are granted access to Claude Mythos Preview to identify and fix vulnerabilities in critical systems. Critics and competitors of Anthropic are not buying the claimed capabilities of Claude Mythos and are demanding for more proof.

As AI models become more capable than ever before—now surpassing all but the most experienced experts in security—the need to ensure that defensive actors retain an edge over malicious actors becomes critical.

Also Read: Claude Code Leak: Human Error or System Failure at Anthropic?