Trump Called Anthropic a 'National Security Menace' —...

Anthropic — the AI company behind Claude — was publicly labeled a "national security menace" by the Trump administration for nearly two months after refusing to build a mass surveillance system. That refusal, and Anthropic's pivot to a defensive cybersecurity AI, is now at the center of a broader battle over what AI automation should and should not do.

When the Government Calls Your AI Company a National Threat

The White House feud with Anthropic didn't start quietly. Administration officials publicly labeled the company behind Claude a national security risk — language typically reserved for foreign adversaries, not California AI labs. The attacks ran for roughly two months, with tensions hitting rock bottom in late February 2026 when the Pentagon-Anthropic relationship reportedly collapsed.

Behind the political noise, a clearer picture has emerged: Anthropic drew two hard red lines — two specific capabilities it would not build regardless of who was asking:

• Domestic mass surveillance — using Claude to monitor, track, or profile U.S. citizens at scale inside the country
• Lethal fully autonomous weapons — AI systems that select and engage targets without a human making the final decision (sometimes called "killer robots" in arms control policy discussions; these differ from remote-controlled drones because no human approves each individual strike)

Neither is a hypothetical. The U.S. military and intelligence community have been actively exploring AI-assisted surveillance systems for years. Fully autonomous lethal drones — systems that identify, track, and strike without any human in the decision loop — are already deployed by several nations. Anthropic's public refusal to build either tool put it on a direct collision course with an administration that was, at minimum, exploring both options.

Anthropic's Cybersecurity AI Pivot: Claude Mythos as a Government Olive Branch

In April 2026, Anthropic launched Claude Mythos Preview — a cybersecurity-focused AI model (a specialized version of Claude trained to identify cyber threats, analyze software vulnerabilities, and support digital defense teams — think of it as Claude coached to think like a security analyst rather than a general assistant). The launch appears strategically calculated: rather than simply refusing government work, Anthropic is offering an alternative path — helping defend systems rather than surveilling people.

Political tensions that had been near a boiling point appear to be thawing. The strategic logic is clear: a cybersecurity AI that protects government infrastructure from foreign hackers is a fundamentally different product than a domestic monitoring tool. Anthropic is betting that demonstrating defensive value can extract it from two months of political crossfire — and potentially restore a Pentagon relationship that deteriorated sharply in late February 2026.

The business stakes are significant. Claude models power thousands of commercial applications, and a prolonged government blacklist would cut Anthropic off from federal AI contracts — a market worth billions of dollars annually. Claude Mythos Preview is Anthropic's opening bid for a seat at that table, without surrendering either of its two hard red lines. Developers evaluating which AI automation platform aligns with their compliance needs can explore our AI platform comparison guides.

While Altman Fights in Washington, His Other Company Wants Your Iris

Sam Altman — CEO of OpenAI — also runs a separate company called World (formerly Worldcoin). While OpenAI wages competitive battles in the AI model market, World just expanded its most controversial product to the United States: an iris-scanning orb (a physical device roughly the size of a bowling ball that photographs your face and eyes to create a permanent, unique biometric identifier — like a fingerprint scan, except you cannot change your irises if a breach ever happens).

The incentive structure is simple by design. Tinder users in Japan and now the United States can visit a World orb station, submit to facial and iris photographs, and receive 5 free Tinder boosts (a premium feature that dramatically increases your profile's visibility to other users, normally priced at several dollars each) in exchange for permanent biometric access. World encrypts and stores the data.

Several things deserve close attention:

• Iris scans are permanent biometric data — unlike passwords or card numbers, you cannot change your eyes if a database is ever compromised
• World's encryption claims have not been independently audited or publicly verified by a neutral third party
• The reward structure asks users to trade lifetime biometric access for roughly $10–$15 worth of app perks (5 Tinder boosts)
• World first piloted the Tinder orb program in Japan last year, before rolling out to both Japan and the United States in 2026, building toward what Altman describes as a global identity layer

The deeper concern isn't whether the data is currently secure — it's whether concentrating permanent biometric identifiers from millions of users inside a single private company creates a systemic risk that no encryption scheme can fully eliminate. World is building quietly, one market and one app partnership at a time.

OpenAI Abandons Sora Video AI — Key Researcher Exits

Sora — OpenAI's video generation system (an AI tool capable of producing realistic video clips from written text descriptions, which OpenAI had marketed as a breakthrough in creative AI) — has been effectively abandoned. The company is redirecting resources away from what internal sources describe as research "side quests" toward higher-margin products: coding tools and enterprise software.

Bill Peebles, the senior researcher who led the Sora team, announced his departure from OpenAI alongside this strategic shift. In his farewell, Peebles wrote: "It's tempting in life to mode collapse to the most immediate priorities." For anyone watching closely, that's a pointed observation from a research lead about a company pivoting hard toward quarterly revenue over long-horizon bets.

The 2026 trajectories of the three major players paint sharply different pictures:

• OpenAI — cutting research moonshots, losing senior team leads, doubling down on Codex (an AI coding system capable of controlling desktop applications and running multiple automated tasks simultaneously) and enterprise sales
• Anthropic — absorbing roughly 2 months of political attacks, refusing surveillance contracts, launching targeted government-compatible products (Claude Mythos), and potentially rebuilding Pentagon relationships
• World (Altman's separate company) — expanding biometric identity infrastructure across markets with limited regulatory scrutiny, monetizing through consumer app partnerships including Tinder in Japan and the United States

Three AI Companies. Three Very Different Visions of AI Automation.

What this week's converging headlines reveal is a fundamental divergence in what the most powerful AI companies believe the technology should ultimately do — and who should control it.

OpenAI is betting that enterprise coding tools and ChatGPT consumer growth will sustain the company as expensive research projects get cut. Anthropic is betting that principled limits — backed by a new cybersecurity AI — will earn government credibility without requiring it to become a surveillance contractor. And Sam Altman, through World, is quietly building a biometric identity system that may one day become the default way the internet verifies who you are.

If you're a developer choosing which AI automation platform to build on, or a consumer thinking about which apps to trust, these diverging values matter more right now than any benchmark score. Anthropic's refusal on surveillance and autonomous weapons is a policy you're implicitly aligning with when you integrate Claude. OpenAI's enterprise pivot signals where its engineering budget will concentrate. And World's orb expansion is worth watching closely before it shows up in an app you already use. Start with our AI platform comparison guides to see which tools align with your workflow and values.

Related Content — Get Started | Guides | More News