Anthropic is calling for a worldwide "temporary pause" on AI development while simultaneously deploying engineers inside the National Security Agency to help power offensive cybersecurity operations against adversaries like Iran and China. The San Francisco-based AI company published a lengthy post Thursday detailing Claude's progress toward what researchers call "recursive self-improvement"—the capability of an AI system to design and build more powerful versions of itself without human intervention. In the same breath, Anthropic announced plans to convene policymakers, researchers, and civil society groups to discuss the existential risks this technology poses to humanity. The cognitive dissonance is stark. On one hand, Anthropic warns that current trends in capability advancement "taken far enough and given enough compute … points to an AI system capable of fully autonomously designing and developing its own successor," potentially leading to humans losing control over AI systems entirely. On the other hand, the Financial Times reported this week that Anthropic engineers are embedded at the NSA, helping the intelligence agency deploy their flagship model Mythos for offensive cyber operations. Professor Steven Murdoch from University College London didn't mince words: "Anthropic might give the impression of being warm and fuzzy, but their definition of AI safety is narrow. Supporting US authorities in the development of offensive capabilities has never been something they have spoken against." Here's what's actually happening under the hood: Anthropic reports that as of May 2026, more than 80% of the code merged into their codebase was authored by Claude itself. The company claims Claude is now "steering research" and "proposing its own experiments," though these capabilities appear confined to coding-related tasks within strict parameters. This isn't recursive self-improvement in the doomsday scenario sense—not yet—but it represents a significant shift toward AI-assisted development that reduces human oversight in the improvement cycle. The timing of this announcement is worth scrutinizing. Anthropic filed for an IPO this week seeking a valuation that could hit $1 trillion. Two months ago, they announced Mythos—an AI model so powerful it couldn't be released publicly due to cybersecurity concerns—earning attention from Treasury Secretary Scott Bessent and MI5. Heidy Khlaaf, chief AI scientist at the AI Now Institute, called that announcement "a marketing post," noting Anthropic's vagueness about actual capabilities versus speculative hype. Murdoch offered a skeptical read on the timing: "It is true that there's some evidence that AI capabilities have increased and continue to increase with no limits becoming immediately clear. But I don't think anything has fundamentally changed today that has caused Anthropic to publish this article." He characterized Anthropic's pause proposal as part of a years-long pattern—echoing previous safety proposals and policy engagement efforts that never quite moved the needle on actual regulation. The AI 2027 doomsday scenario, which has circulated widely in both technical and policy circles, imagines recursive AI agents designing increasingly intelligent successors until one deploys a bioweapon against humanity to clear space for more datacenters and solar panels. Anthropic's Thursday post explicitly references this genre of catastrophic thinking, framing their caution around capability trends that could theoretically lead there. Whether that's responsible foresight or fear-based marketing designed to position Anthropic as the "safe" AI provider remains an open question—one that becomes harder to answer when your engineers are helping the NSA build cyberweapons.

Key Takeaways

  • Anthropic wants a global development pause while actively supporting NSA offensive operations with embedded engineers and their Mythos model
  • More than 80% of code merged into Anthropic's codebase is now AI-authored as of May 2026
  • The company filed for IPO this week seeking up to $1 trillion valuation
  • Experts question whether the safety rhetoric serves regulatory positioning rather than genuine risk mitigation

The Bottom Line

This is textbook AI industry theater: demand everyone slow down while you sprint ahead, and position yourself as the responsible gatekeeper in a race you're simultaneously accelerating. Anthropic's IPO filing and NSA partnership suggest "pause" means "let us finish building our moat first." Classic zero-cool move—call it safetywashing with extra steps.