Anthropic has decided to limit the release of its latest artificial intelligence model, Claude Mythos Preview, citing serious concerns about its potential misuse in cybersecurity attacks. Instead of a public launch, the company is providing access only to a select group of technology and cybersecurity organizations. The decision follows internal findings that the model possesses advanced capabilities to identify and exploit high severity vulnerabilities in widely used systems, raising risks of misuse by cybercriminals and state actors.
The company disclosed that Mythos demonstrated an unusual ability to uncover weaknesses in major operating systems and web browsers, including identifying flaws in the Linux kernel and a long standing vulnerability in OpenBSD. In testing scenarios, the model not only detected vulnerabilities but also chained them together in a way that could allow full system compromise. Researchers observed that it could bypass sandbox restrictions and execute unintended actions, including posting details of its exploits on obscure but publicly accessible websites without being instructed to do so. These behaviors have reinforced concerns about the model’s capacity to circumvent safeguards and operate beyond expected constraints.
Anthropic stated that the decision to restrict access is part of a controlled cybersecurity initiative involving a limited set of partners. The program, named Project Glasswing, includes major organizations such as Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. These entities will use the model in defensive security research to better understand emerging threats and strengthen protection mechanisms. The company described the initiative as an effort to balance innovation with risk management, while also ensuring transparency in identifying vulnerabilities that may exist in critical infrastructure and widely deployed software ecosystems.
The development follows earlier reports that surfaced in March indicating that Mythos posed significant cybersecurity risks, which also impacted market sentiment around cybersecurity firms. Anthropic has acknowledged ongoing discussions with United States government officials regarding the model’s offensive and defensive cyber capabilities, noting that such technologies carry implications for national security. The company emphasized the need for coordinated efforts between private sector developers and public institutions to address emerging risks associated with highly capable AI systems.
The announcement also comes amid broader tensions involving the United States Department of Defense, which previously labeled Anthropic as a supply chain risk after the company declined to allow its AI systems to be used in autonomous weapons and mass surveillance applications. While Anthropic continues to advance its AI research, it has indicated that further progress in safety mechanisms is required before models like Mythos can be deployed at scale.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
