AI Pulse by Inblix

Anthropic drops Claude Opus 5 with tougher cyber defenses after Fable 5 standoff

The Verge AI · Jul 24, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: Anthropic drops Claude Opus 5 with tougher cyber defenses after Fable 5 standoff

Anthropic just shipped Claude Opus 5, and the timing is everything. It lands weeks after the company’s Fable 5 model got yanked offline and stuffed with new safeguards following a tense back-and-forth with the US government. Opus 5 is being sold as the safer, more compliant sibling — it’s the one you use when you can’t afford a national security headache.

The company says Opus 5 “comes close to the capabilities of Claude Fable 5 in many domains” and is much sharper on complex coding tasks. But the real headline is the security posture. Anthropic claims this is “the most aligned Opus model” and the hardest to trick into misuse. They’re clearly reading the room after Fable 5 kicked off a new, more paranoid era of AI regulation under the Trump administration. Spokesperson Danielle Ghiglieri confirmed the company is working with government partners on independent testing of Opus 5 — a process that’s rapidly becoming the new normal, just as OpenAI’s GPT-5.6 went through a two-week government-only preview.

Pricing is also a flex. At $5 per million input tokens and $25 for output, Opus 5 matches its predecessor’s cost and clocks in at “half the price” of the more powerful Fable 5. It’s slightly cheaper than GPT-5.6 too. That feels like a pointed response to the industry’s “AI money squeeze,” where labs are finding creative new ways to pass costs to users. Anthropic even rolled out a “Fast mode” for double the price, plus an automatic fallback option that bumps you to a weaker model if the safeguards kick in — a feature that’s as much about risk management as it is about user experience.

The subtext here is that enterprise buyers are getting a model that’s deliberately less capable than Fable 5 in some respects, but compliant enough to actually deploy without lawyers panicking. For knowledge work and biology, Anthropic says it’s the ideal daily driver. Whether customers agree that trading peak performance for political safety is a good deal remains an open question.

💡 Key Takeaways

  1. Opus 5 was built to placate government concerns, not push the frontier — it's deliberately less capable than the temporarily banned Fable 5.
  2. At $25 per million output tokens, Anthropic is keeping Opus 5's price flat while undercutting both its own Fable 5 and OpenAI's GPT-5.6
  3. The new automatic fallback feature means users might silently get shifted to a weaker model when safeguards kick in, raising transparency questions.
  4. Government pre-release testing, once an anomaly, is now standard procedure for frontier model launches at both Anthropic and OpenAI.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles