Anthropic says its powerful Claude Mythos class AI models could soon become available to all customers after months of restricted testing and safety evaluations.
Quick Summary – TLDR:
- Anthropic plans to roll out Mythos class models to all customers in the coming weeks.
- The company initially restricted Mythos due to concerns about its advanced cybersecurity capabilities.
- Mythos has identified more than 23,000 software vulnerabilities and helped uncover critical flaws in major software projects.
- Anthropic says it is developing safeguards to reduce misuse before a broader release.
What Happened?
Anthropic has confirmed that it is preparing to bring its highly anticipated Claude Mythos class AI models to the public. The company says it has made significant progress on safety protections that were previously considered necessary before wider access could be granted.
The announcement came as part of the launch of Claude Opus 4.8, where Anthropic revealed that Mythos class models are expected to reach customers in the coming weeks, although it has not provided a specific release date.
Claude Mythos is launching in June.
— BridgeMind (@bridgemindai) May 28, 2026
Anthropic just confirmed it.
“We expect to bring Mythos-class models to all our customers in the coming weeks.”
A new class of model with even higher intelligence than Opus. Currently in preview with select organizations for cybersecurity… pic.twitter.com/EXVthXastJ
Why Anthropic Delayed Mythos?
When Anthropic first introduced Claude Mythos in April, the company chose not to release it broadly. Instead, access was limited to selected organizations and security researchers through its Project Glasswing initiative.
The decision reflected concerns about the model’s ability to discover software vulnerabilities and generate working exploits. Anthropic warned at the time that powerful AI systems could temporarily give attackers an advantage if proper safeguards were not in place.
The company argued that while defenders would ultimately benefit from advanced cybersecurity AI, an uncontrolled release could create short term risks for software developers, businesses, and critical infrastructure operators.
What Makes Mythos Different?
According to Anthropic, Mythos delivers substantial improvements in code reasoning, autonomy, and cybersecurity analysis compared with existing Claude models.
During internal testing, Mythos reportedly generated working software exploits 72.4 percent of the time, while earlier Claude models achieved little to no success in similar tests.
Researchers also reported that Mythos could autonomously search for software vulnerabilities and produce functional exploit chains with minimal human guidance. These capabilities quickly elevated the model into a category of AI systems that security experts view as both highly useful and potentially risky.
Anthropic has described Mythos as significantly more capable than current flagship models, including Claude Opus 4.8.
Thousands of Vulnerabilities Already Found
The company says Mythos has already demonstrated its effectiveness in real world security testing.
According to Anthropic, the model has scanned more than 1,000 open source projects and identified over 23,000 vulnerabilities, including more than 6,000 classified as high or critical severity.
One notable discovery involved a vulnerability in the wolfSSL cryptography library. The flaw could have allowed attackers to forge certificates and impersonate legitimate websites before it was patched.
Anthropic also said Mythos Preview has uncovered vulnerabilities affecting major operating systems and web browsers.
Project Glasswing Partners See Major Results
Through Project Glasswing, Anthropic partnered with major technology companies including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, and Google to evaluate Mythos in controlled environments.
The company claims that most participating organizations discovered hundreds of serious vulnerabilities within their own software systems using the model.
Among the most notable examples was Cloudflare, which reportedly found around 2,000 bugs across critical systems. Nearly 400 of those vulnerabilities were categorized as high or critical severity issues.
Anthropic also said Cloudflare found Mythos generated fewer false positives than many human security testers.
Several Project Glasswing participants reportedly experienced more than a tenfold increase in their bug discovery rates after adopting the model.
Safety Remains the Key Focus
Despite the impressive results, Anthropic continues to emphasize that strong safeguards remain essential.
The company acknowledged that no organization has yet developed perfect protections against misuse of advanced cybersecurity AI. Before a full rollout, Anthropic plans to continue refining mechanisms designed to prevent malicious use while preserving the model’s usefulness for defenders.
The company is also expanding access to certain government agencies and allied organizations before broader public availability.
In a recent update, Anthropic stated:
SQ Magazine Takeaway
I think Claude Mythos could become one of the most important AI releases in cybersecurity. The numbers reported by Anthropic are impressive, particularly the discovery of thousands of serious vulnerabilities that might otherwise have remained hidden. At the same time, the company’s caution highlights a growing challenge facing the AI industry.
As models become more capable of finding and exploiting software flaws, balancing innovation with security will become increasingly important. The public release of Mythos could mark a major turning point in how organizations defend software systems against future threats.
