Fearing Hackers, Anthropic Delays Release of new Powerful AI Tool Claude Mythos

Anthropic has developed a powerful new artificial intelligence model that experts say could transform cybersecurity. The company has chosen not to release it to the public. This decision comes amid concerns that the technology might allow hackers to launch attacks far more quickly and effectively than before.

The model, called Claude Mythos Preview, represents a significant leap in AI capabilities. It excels at spotting and exploiting software weaknesses. However, Anthropic has restricted access to a small group of trusted partners. This move aims to strengthen defenses before the model reaches potential adversaries.

Claude Mythos A Striking Leap in Vulnerability Detection

Claude Mythos Preview has already identified thousands of high-severity vulnerabilities. These include zero-day flaws in every major operating system and web browser. For instance, the model uncovered a 27-year-old bug in OpenBSD that could cause remote crashes. It also found a 16-year-old issue in the FFmpeg video codec.

Moreover, the AI can autonomously create working exploits. It has demonstrated this on complex targets, such as chaining vulnerabilities in the Linux kernel to gain higher privileges. In one case, it exploited a remote code execution flaw in FreeBSD to achieve full system control. This level of autonomy goes well beyond what previous models could achieve.

Anthropic announced the preview on April 7, 2026. Researchers noted that the model generates functional exploits in hours. Human experts often need days or weeks for similar tasks. This has alarmed many in the field.

Project Glasswing Brings Partners Into the Fold

Anthropic has launched an initiative named Project Glasswing. Through it, the company shares limited access to Claude Mythos Preview with key industry players. Partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks.

These organizations use the model to scan for bugs and test hacking techniques. Their goal is to patch critical software before flaws become public. This collaborative effort targets essential systems that power the internet, banking and other vital infrastructure.

Anthropic has also pledged resources to support open-source security. The company is providing $100 million in usage credits and $4 million in direct donations. Public reports on findings and fixes are expected within 90 days.

Experts Highlight the Risk of Claude Mythos

Cybersecurity specialists have expressed mixed views on the development. Some warn that a single hacker equipped with this technology could try thousands of attack variations rapidly. This shift could make sophisticated exploits more common and harder to stop.

The model also shines at creating exploit chains. These sequences combine multiple weaknesses to breach systems deeply. As a result, attacks could become faster and more persistent than those mounted by teams of human experts today.

However, not all concerns have reached the level of panic. Defenders still lack comparable AI tools in many cases. Earlier Anthropic models have inspired some defensive applications, yet the gap remains wide. This imbalance has prompted calls for broader investment in AI-driven security.

Anthropic Stresses Caution and Collaboration

Anthropic has made its position clear. The firm will not release Claude Mythos Preview generally available. Officials cite the high risk of misuse by cybercriminals or state actors. Most of the vulnerabilities the model discovered remain unpatched.

Instead, the company has briefed senior U.S. government officials on the technology. It has offered support for official testing and evaluation. This approach seeks to level the playing field by arming defenders first.

Logan Graham, who leads Anthropic’s team focused on AI safeguards, has emphasized the need for careful handling. The company continues to develop protections for future models. These include systems to detect and block dangerous outputs.

Web Resources on Anthropic Delays Release of Claude Mythos

1. CNN.com : Anthropic’s latest AI model could let hackers carry out attacks faster
2. CNBC.com :Anthropic rolls out Claude Opus 4.7, an AI model that is less risky
3. Financial Times ; Anthropic rolls out cyber AI model days after source code leak