Anthropic has stated that its newest upcoming AI mannequin outperforms each different massive language mannequin (LLM) it has launched thus far, suggesting a significant potential leap within the race to develop cutting-edge frontier AI techniques.
The unreleased mannequin marks a “step change” in AI efficiency and is “probably the most succesful we’ve constructed thus far,” an Anthropic spokesperson was quoted as saying by Fortune. The excitement across the mannequin comes after an inside knowledge leak uncovered its existence to cybersecurity researchers Roy Paz and Alexandre Pauwels, who found a draft announcement weblog submit containing an outline of the mannequin and different particulars in an unsecured and publicly searchable knowledge cache.
The leak has additionally been confirmed by Anthropic, which reportedly acknowledged {that a} “human error” within the configuration of its content material administration system (CMS) led to the draft weblog submit being accessible. Whereas Anthropic has but to say when it plans to unveil the brand new mannequin, right here’s what we all know to this point.
Key particulars about Anthropic’s next-gen AI mannequin
The brand new mannequin described within the leaked weblog submit is known as ‘Claude Mythos’. Anthropic has stated that it has accomplished coaching Mythos and has opened up entry to the mannequin to a small group of early entry customers.
Mythos is reportedly a part of a completely new lineup of AI fashions by Anthropic referred to as ‘Capybara’. Presently, Anthropic gives every of its AI fashions in three totally different sizes, specifically: Opus, which is the biggest and most succesful mannequin variations; Sonnet, cheaper, barely sooner but in addition much less succesful; and Haiku, its most cost-effective, smallest, and quickest model of LLMs.
“‘Capybara’ is a brand new title for a brand new tier of mannequin: bigger and extra clever than our Opus fashions—which had been, till now, our strongest,” Anthropic has been quoted as saying within the weblog submit. Fashions on this new tier are reportedly bigger and extra succesful than Opus, but in addition dearer to make use of.
Mythos’ capabilities reportedly pose unprecedented cybersecurity dangers, which has Anthropic involved about its real-world implications. “In comparison with our earlier greatest mannequin, Claude Opus 4.6, Capybara will get dramatically increased scores on exams of software program coding, tutorial reasoning, and cybersecurity, amongst others,” the corporate was quoted as saying within the leaked materials.
Story continues under this advert
“In making ready to launch Claude Capybara, we wish to act with further warning and perceive the dangers it poses—even past what we study in our personal testing. Specifically, we wish to perceive the mannequin’s potential near-term dangers within the realm of cybersecurity—and share the outcomes to assist cyber defenders put together,” the doc learn.
Certainly one of Anthropic’s considerations is that the mannequin may very well be misused by risk actors to run large-scale cyber assaults. Mythos is “at present far forward of another AI mannequin in cyber capabilities,” and “presages an upcoming wave of fashions that may exploit vulnerabilities in ways in which far outpace the efforts of defenders,” the corporate notes within the weblog submit.
These considerations may clarify Anthropic’s cautious rollout technique of Mythos, beginning with a small group of early-access customers. The mannequin can be costly to run and never but prepared for basic launch, as per the draft weblog submit. When it’s launched extra broadly, Anthropic is predicted to make it first accessible to cyber defenders. “We’re releasing it in early entry to organizations, giving them a head begin in enhancing the robustness of their codebases in opposition to the upcoming wave of AI-driven exploits,” it was quoted as saying.
AI disruption of cybersecurity business
Even earlier than its official launch, Mythos’ reportedly superior cyber capabilities sparked a sell-off of cybersecurity shares on Friday, March 27, in accordance with a report by CNBC. Shares of iShares Cybersecurity ETF fell 4.5 per cent, whereas CrowdStrike, Palo Alto Networks, Zscaler, and SentinelOne shares declined by 6 per cent every. Tenable inventory noticed the sharpest fall of 9 per cent.
Story continues under this advert
Cybersecurity firms are beneath immense stress to evolve together with a shifting risk panorama spurred by the rise of AI-powered instruments and autonomous brokers, which have considerably lowered the entry barrier to hold out refined cyber assaults. In November 2025, Anthropic stated {that a} state-sponsored group in China used its Claude AI mannequin to automate a cyber assault. Stories of Anthropic testing Mythos additionally come days after The Data reported that archrival OpenAI has completed pre-training a brand new, very sturdy mannequin known as ‘Spud’.