Anthropic lifted the curtain on two major developments Thursday: a new release of its Claude Opus generative AI model and a near‑term plan to open its Mythos‑Preview system to the wider market. The company’s blog described Opus 4.8 as a "modest but tangible improvement" over the previous 4.7 version, citing better benchmark scores, reduced instances of unsupported claims, and more frequent expressions of uncertainty.
Beyond the incremental upgrade, Anthropic signaled that the Mythos‑Preview model, currently sandboxed for a select group of cybersecurity firms and tech partners, will soon be offered to all customers. The model resides in a partnership program called Project Glasswing, which Anthropic says is designed to give experts a "lead time to patch flaws" the AI uncovers. "Models of this capability level require stronger cyber safeguards before they can be generally released," the company wrote, adding that it expects to roll out Mythos‑class models to the broader user base in the coming weeks.
Security researchers have already put Mythos to the test. In one notable case, Mozilla’s latest Firefox release incorporated more than 200 fixes that Mythos identified. The model’s ability to locate exploits faster than human hackers has drawn both praise and caution. Darren Williams, founder and CEO of cybersecurity firm BlackFog, called the staged release "the right instincts" but warned that "the window between a powerful model's release and broad adoption of defenses against it is always a vulnerable moment."
Cost is another factor shaping the model’s impact. Jake Williams, a researcher at IANS Research, said internal tests showed Mythos costing roughly thirty times more to run than the prior Opus model. "This is outside the reach of many, including any commodity threat actors," he told CNET. "Nation‑state actors already had better technology for finding vulnerabilities. This is only changing the game for a small percentage of threat actors."
Anthropic also introduced a new "effort" control feature for its Claude.ai and Claude Cowork products. Users can now tell the model how much computational effort to expend on a response. Higher effort settings promise more nuanced answers but will deplete usage limits faster, while lower settings deliver quicker, more economical replies.
The dual announcements arrive amid a broader AI arms race, with competitors such as OpenAI and DeepSeek also unveiling new models. Anthropic’s emphasis on cybersecurity safeguards reflects growing industry concerns about the dual‑use nature of powerful generative AI. By limiting early access to trusted partners, the company hopes to iron out vulnerabilities before the technology reaches a mass audience.
While the rollout of Opus 4.8 offers incremental gains for everyday users, the forthcoming public availability of Mythos could reshape how both defenders and attackers leverage AI. The balance between innovation and risk mitigation will likely define Anthropic’s next few weeks as it moves from a controlled preview to a full‑scale release.
This article was written with the assistance of AI.
News Factory SEO helps you automate news content for your site.