Anthropic announced Thursday it expects to expand access to its Claude Mythos AI model "in the coming weeks" after completing additional safeguards testing. The company made the statement alongside its new Opus 4.8 model release, marking its clearest indication yet of wider availability for the cybersecurity-focused system. Anthropic has restricted Mythos access through Project Glasswing since March, following warnings from researchers and government agencies about the model's autonomous cyberattack capabilities. The announcement comes amid broader AI industry debate over how advanced systems with offensive security capabilities should be released to customers.
Current Access Restrictions
Anthropic has limited Claude Mythos distribution through Project Glasswing, a program providing select technology companies, security researchers, and government partners access under controlled conditions. The company first described Mythos in leaked March blog materials as "the most powerful AI model we've ever developed," positioning it as a tier above its Opus models. Anthropic stated the model "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." The company did not specify which safeguards remain incomplete before wider rollout or whether all customers will receive identical access levels. Anthropic did not respond to requests for comment on implementation details.
Documented Technical Capabilities
The U.K.'s AI Security Institute found Mythos autonomously completed a 32-step simulated corporate network attack during testing. In April, Mozilla reported Mythos identified 271 vulnerabilities in Firefox during internal security evaluations. Earlier this month, security startup Calif stated a preview version helped researchers develop an exploit chain targeting Apple's M5 Mac chips. Anthropic has argued Mythos could help defenders identify and patch software vulnerabilities before attackers exploit them. Security researchers and government agencies warned the same capabilities could accelerate cyberattacks.
Industry Debate Over Release Strategy
The model has become central to discussions inside the AI industry over advanced system releases. During a May interview with the "Core Memory" podcast, OpenAI CEO Sam Altman accused Anthropic of using "fear-based marketing," arguing warnings about cybersecurity risks could justify limiting access to powerful AI systems. Users on Myriad—a prediction market platform operated by Decrypt's parent company Dastan—gave a 44% probability to Mythos releasing by end of June as of Thursday, up from 17.5% that morning.
Confirmed Next Steps
Anthropic stated it is "making swift progress on developing these safeguards" and expects to bring "Mythos-class models to all our customers in the coming weeks."
FAQ
What is Project Glasswing?
Project Glasswing is Anthropic's restricted access program providing select technology companies, security researchers, and government partners controlled access to the Claude Mythos model under specific conditions.
What vulnerabilities has Mythos identified in testing?
Mozilla reported Mythos identified 271 vulnerabilities in Firefox during April internal security evaluations. The U.K.'s AI Security Institute documented the model autonomously completing a 32-step simulated corporate network attack. Security startup Calif stated a preview version helped develop an exploit chain targeting Apple's M5 Mac chips.
When did Anthropic first announce Mythos?
Claude Mythos first surfaced in March after draft Anthropic blog materials leaked online. The company described it as "the most powerful AI model we've ever developed" and positioned it as a new tier above its Opus models.
