Artificial intelligence company Anthropic announced Friday it will not release its latest AI model to the general public after determining the technology poses significant security risks that could be exploited by malicious actors.
The model, called Claude Mythos Preview, will instead be distributed exclusively to a select group of major technology companies including Microsoft, Apple, CrowdStrike and Amazon Web Services, along with more than 40 organizations responsible for building critical software infrastructure worldwide.
The restricted rollout forms part of Project Glasswing, a new cybersecurity initiative aimed at identifying and patching vulnerabilities in essential software systems before they can be exploited by hackers or foreign adversaries.
Unprecedented Vulnerability Detection Capabilities
According to Anthropic's internal testing, Claude Mythos Preview has already identified thousands of high-severity security flaws across major computing platforms. The discoveries include previously unknown vulnerabilities in widely-used operating systems and web browsers that had remained undetected for more than two decades.
The company's security researchers found that the AI system's advanced analytical capabilities allow it to scan code and identify potential weaknesses at a speed and scale far beyond traditional security auditing methods.
"The model demonstrates an ability to discover complex, multi-step vulnerabilities that require deep understanding of software architecture and potential attack vectors," Anthropic stated in its announcement.
Dual-Use Technology Concerns
While the AI system's vulnerability detection capabilities offer significant benefits for cybersecurity defense, Anthropic warned that the same technology could be weaponized by malicious actors to identify and exploit security flaws.
The company expressed particular concern about the potential for the technology to fall into the hands of cybercriminal organizations or nation-state actors who could use it to launch sophisticated attacks against critical infrastructure, government systems, or private networks.
"Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely," Anthropic said in its statement. "The fallout — for economics, public safety, and national security — could be severe."
Project Glasswing Partnership Framework
Under the Project Glasswing initiative, participating organizations will use Claude Mythos Preview exclusively for defensive cybersecurity purposes. Partner companies that build or maintain critical software infrastructure will employ the model to scan both proprietary and open-source systems for potential vulnerabilities.
The project includes a framework for sharing vulnerability discoveries across the technology industry while ensuring that sensitive information about security flaws is handled responsibly and disclosed through appropriate channels.
Anthropic said the initiative represents a collaborative approach to addressing cybersecurity challenges in an era where AI capabilities are rapidly advancing and becoming more widely accessible.
Financial Commitment to Open-Source Security
As part of Project Glasswing, Anthropic announced it is committing up to $100 million in usage credits to allow participating organizations to access and utilize Claude Mythos Preview for security research and vulnerability detection.
Additionally, the company pledged $4 million in direct donations to open-source security organizations that work to identify and patch vulnerabilities in widely-used software libraries and frameworks.
The financial commitment aims to strengthen the overall security ecosystem by providing resources to organizations that may lack the budget to access advanced AI-powered security tools independently.
Industry Response and Implications
The decision to restrict access to Claude Mythos Preview marks a significant departure from the traditional approach of many AI companies, which have generally favored broader public releases of their models.
Cybersecurity experts have praised Anthropic's cautious approach, noting that the proliferation of AI-powered hacking tools could create an unprecedented threat landscape if not properly managed.
The move also highlights growing concerns within the technology industry about the potential misuse of advanced AI systems for malicious purposes, particularly in the cybersecurity domain where the stakes for critical infrastructure protection are exceptionally high.
Future Security Landscape Transformation
Despite the security risks, Anthropic emphasized that AI-powered vulnerability detection technology offers substantial benefits for improving software security across the technology industry.
The company noted that AI systems like Claude Mythos Preview can help developers identify and fix security flaws much earlier in the software development process, potentially leading to more secure applications and systems reaching end users.
Anthropic described Project Glasswing as a crucial step toward ensuring that defenders maintain an advantage in what the company characterized as the emerging AI-driven era of cybersecurity, where both attackers and defenders will increasingly rely on artificial intelligence tools.
The initiative is expected to generate valuable insights about how AI capabilities can be responsibly deployed in cybersecurity contexts while minimizing risks to global digital infrastructure.
Photo credit: Photo courtesy of Anthropic