Anthropic vs. Washington: No Safe Mode

AI Unpacked

Welcome back to AI Unpacked, your weekly briefing on the biggest developments shaping how artificial intelligence is moving from labs into the real world.

This week, the spotlight turns to the most dramatic rupture yet between Silicon Valley and Washington — a high-stakes showdown between AI safety and military power that ended with a government ban, a rushed rival deal, and a stunning irony: the banned AI was still being used in combat while the ink on the ban was barely dry.

Here's what mattered most.

The Anthropic-Pentagon Blowup: A Rupture Years in the Making

The conflict had been building for months, but it came to a head fast.

The standoff was reportedly set off by the U.S. military's use of Claude during the operation to capture former Venezuelan President Nicolás Maduro in January. CBS News From there, tensions escalated quickly. Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei until the end of the week to provide a signed document granting the military full access to its AI model. Officials were also considering invoking the Defense Production Act to compel compliance. CBS News

At the center of the dispute were two guardrails Anthropic refused to remove. The company maintained that AI-driven mass surveillance presents serious, novel risks to fundamental liberties, and that today's frontier AI systems are simply not reliable enough to power fully autonomous weapons — meaning they could endanger America's warfighters and civilians. Anthropic

The Pentagon's position was equally firm. Officials argued that Anthropic's restrictions could create a dynamic where the military gets used to how a model works, and then when an urgent situation arises, it gets blocked from using it. CBS News

The result was a total breakdown. After negotiations failed, President Trump directed federal agencies to stop using Anthropic's technology, while Secretary Hegseth designated the company a "supply-chain risk to national security" — a label previously reserved for foreign adversaries, never before applied to an American company. NPR

The central tension is not going away. AI is now deeply embedded in military operations. The question of who controls how it's used — the government or the company that built it — will define the next era of AI governance.

Sam Altman Steps In — and Steps Into the Deal

When Anthropic was blacklisted, OpenAI moved fast. But Sam Altman tried, at first, to play peacemaker.

In a memo to OpenAI staff, Altman wrote that the dispute risked national security and American leadership in AI, and that OpenAI wanted to help de-escalate things. CNN He signaled that OpenAI shared Anthropic's core red lines. Altman wrote that OpenAI has long believed AI should not be used for mass surveillance or autonomous lethal weapons, and that humans should remain in the loop for high-stakes automated decisions. CNBC

But de-escalation gave way to dealmaking. Altman announced that OpenAI reached an agreement allowing the Department of Defense to use its AI models on classified networks, with technical safeguards to ensure the models behave as intended, and engineers embedded with the Pentagon to help with safety. TechCrunch

The timing raised eyebrows across the industry. By Altman's own admission, OpenAI's deal with the Defense Department was "definitely rushed," and "the optics don't look good." TechCrunch Critics quickly pointed out that OpenAI had secured essentially the same guardrails that got Anthropic banned — raising uncomfortable questions about whether the outcome was about safety at all, or simply about corporate relationships.

Altman himself acknowledged on X that the government's blacklisting of Anthropic set "an extremely scary precedent," adding: "As the more powerful party, I hold the government more responsible." Yahoo!

The Irony That Defined the Week

The most jaw-dropping moment came not from a boardroom, but from a battlefield.

Hours after President Trump directed every federal agency to immediately cease using Anthropic's technology, Claude was employed during U.S. airstrikes on Iran. CENTCOM in the Middle East deployed Claude to assess intelligence, identify targets, and simulate potential battle plans. Inc

Claude is so tightly integrated into military operations that shutting it off immediately could disrupt ongoing missions — which is why it reportedly remained in use during the Iran strike. International Business Times The government itself had acknowledged this reality, giving the Pentagon a six-month phase-out window even as Trump called for an immediate halt.

The contradiction was hard to ignore: the same AI deemed a national security risk on Friday was helping plan strikes on Iran by Saturday morning.

The Big Picture

What this week exposed is not just a dispute between one company and the Pentagon. It revealed something deeper about where AI governance stands in 2026.

AI is no longer experimental. It is operational — woven into intelligence workflows, military planning, and national security infrastructure in ways that make it nearly impossible to remove on a political timeline. The Anthropic episode shows that the real battle isn't just over contracts. It's over who sets the rules for how the most powerful technology in history gets used — and whether safety-focused companies can hold the line when governments push back.

The next decade of AI won't just be about who builds the smartest models.

It will be about who gets to decide how those models are allowed to act.