The Safe, Sanitized Opus 4.7
Anthropic quietly dropped Claude Opus 4.7 this week, billing it as their most powerful generally available model yet. It’s supposedly better at complex coding, image analysis, and generating creative slides. But the company’s own system card tells a different story. Opus 4.7 fails to advance Anthropic’s capability frontier, scoring worse than the Mythos Preview on every single evaluation. This isn’t an upgrade, it’s a lobotomized release designed to keep the real power under lock and key.
The Mythos Preview Paywall
The real star is Claude Mythos Preview, a cybersecurity powerhouse Anthropic announced earlier this month. But you can’t have it. Only a handful of elite partners like Nvidia, JPMorgan Chase, Google, Apple, and Microsoft get private access. Anthropic claims they are testing cyber safeguards on less capable models first. Opus 4.7 is that guinea pig. Anthropic openly admits they trained Opus 4.7 to deliberately reduce its cybersecurity capabilities. So the public gets a neutered model while corporate giants feast on the cutting edge.
Cyber Verification or Just Another Gate?
Anthropic is offering a Cyber Verification Program for security professionals who want to use Opus 4.7 for vulnerability research, theoretically loosening some of the new safeguards. But this feels like a PR move to deflect criticism of their two tiered access system. If Anthropic is serious about safety, they should be transparent about what Mythos Preview can actually do and why they trust only a select few with it. Otherwise, this looks less like responsible deployment and more like a club for the connected few.
Source: Theverge
