Anthropic Preps Opus 4.7 and Full-Stack AI Studio—Whereas Sitting on One thing A lot Scarier – Decrypt

Anthropic is gearing as much as launch Claude Opus 4.7 alongside a brand new AI-powered design instrument that lets customers construct web sites, shows, and touchdown pages with plain English prompts—information that brought about a dip in Adobe, Wix, and Figma shares on Monday, in keeping with The Data.

The merchandise might drop as quickly as this week, an individual with data of the plans advised The Data. The design instrument targets builders and non-technical customers alike, placing it on a collision course with startups like Gamma and Google’s Sew.

Anthropic didn’t reply to Decrypt’s request for remark.

Opus 4.7 is not even Anthropic’s strongest mannequin. That title belongs to Claude Mythos—a cybersecurity-focused beast the corporate is quietly handing to pick safety corporations whereas retaining it away from the general public.

The UK’s AI Safety Institute just lately evaluated Mythos Preview and located it may well autonomously execute refined cyber assaults at charges no different mannequin has matched. It turned the primary AI to finish “The Final Ones,” a 32-step company community assault simulation that usually takes human purple groups 20 hours. Mythos nailed it in three out of ten makes an attempt, averaging 22 of 32 steps—in comparison with Opus 4.6’s 16.

This issues past enterprise safety. Measuring what AI can truly do has turn into an industry-wide headache. OpenAI just lately known as the main coding benchmark “contaminated,” but fashions proceed to be in contrast utilizing those self same checks. A separate ARC-AGI-3 analysis noticed Gemini rating 0.37% and GPT-5.4 hit 0.26%—whereas people received 100%. The result’s a panorama the place benchmarks are each contested and nonetheless used as proof, making it troublesome to contextualize claims about Opus 4.7’s positive aspects till Anthropic releases an in depth mannequin card.

The connection between Opus and Mythos is nearer than most notice. Anthropic builds its frontier fashions by fine-tuning atop the Opus line—the identical spine powering public Claude merchandise will get stress-tested and hardened into Mythos. Opus 4.7 is the inspiration that ultimately will get the cybersecurity kung fu crushed into it.

Additionally, Anthropic’s efforts have been steering extra in the direction of the event/enterprise use case. The leak of Claude code, the discharge of the talents system and MCP protocol, the give attention to agentic AI and the care on coding benchmarks make this much more obvious. Whereas Anthropic hasn’t formally introduced it, the leaks reinforce the broader shift from LLM supplier to one thing that resembles a full-stack “AI studio” mannequin, the place Claude doesn’t simply generate textual content however builds and deploys full merchandise.