Anthropic is expanding its AI model lineup with the release of Claude Opus 4.7, a new offering the company positions as its most capable generally available model to date — while deliberately keeping its most powerful, and potentially most dangerous, technology off the open market.
The San Francisco-based AI firm says Opus 4.7 delivers meaningful improvements over its predecessor, Claude Opus 4.6, across a range of performance benchmarks including agentic coding, multidisciplinary reasoning, scaled tool use and computer use. For enterprise users and developers, the model is designed to handle complex, real-world workflows more effectively — a direct response to the growing demand for AI that can operate with greater autonomy across business processes.
But what makes this launch notable is not just what Claude Opus 4.7 can do — it’s what it deliberately cannot.
Anthropic has engineered the new model to have reduced cyber capabilities compared to Claude Mythos Preview, the company’s most advanced model, which was rolled out earlier this month to a limited group of companies as part of a new cybersecurity initiative called Project Glasswing. Mythos is not generally available and Anthropic has no near-term plans to change that. The company says it is using Project Glasswing as a controlled environment to study how powerful models behave in real-world cybersecurity contexts before considering any broader release.
With Opus 4.7, Anthropic has embedded safeguards that automatically detect and block requests flagged as prohibited or high-risk cybersecurity uses. The company said it also experimented with training techniques aimed at selectively reducing those capabilities at the model level — not just through filtering after the fact. Security professionals with legitimate use cases can apply through a formal verification program to access those capabilities.
The approach reflects the tightrope Anthropic has walked since its founding in 2021 — building competitive, high-performance AI while maintaining what has become the company’s core differentiator: a reputation for safety-first development. That reputation is now being tested at an entirely new scale.
The launch of Project Glasswing has triggered a wave of high-profile conversations across Washington and Wall Street, with members of the Trump administration, tech executives and bank CEOs meeting to assess what Mythos-class AI capabilities could mean for national security and financial infrastructure. The underlying question — how powerful should a publicly available AI model be — is no longer theoretical.
For investors and enterprises, the practical implications of Opus 4.7 are more immediate. The model is priced identically to Opus 4.6, meaning businesses get a material upgrade at no additional cost. It is available across all Anthropic Claude products, its API and through cloud distribution partners Microsoft, Google and Amazon — giving it broad accessibility across the enterprise ecosystem.
The release also signals something important about where the AI industry is heading. Capability tiers are becoming a deliberate strategic tool. The most powerful models are being gated, studied and selectively deployed — not because they aren’t ready, but because the institutions using them need to be.
For small and mid-cap technology companies building on top of AI infrastructure, the implications are significant. As foundation model providers like Anthropic establish formal verification programs and tiered access structures, third-party developers and SaaS companies will need to navigate an increasingly credentialed ecosystem — one where access to the most powerful tools requires demonstrating not just technical fit, but responsible use.
