Guns are fine. Safeties are bad.

Guns are a pretty divisive issue. In America, we have Second Amendment voters who base their ballots on their love of them. Those folks have their counterparts who advocate for legislation and processes around someone collecting firearms like beanie babies. Even with the ideological chasm between them, I bet you could get them to unite if say… the Department of Defense demanded that firearms not have safety mechanisms.

My friends, the Department of Defense is demanding the removal of the safeties.

The Pentagon is threatening to sever its contract with Anthropic and declare the company a "supply chain risk" because it's unwilling to lift certain restrictions on Claude. To wit, Anthropic counters with "Well, you already surprised us by using Claude during the Maduro raid."

The road so far

It's been a lot and moving fast. Dr Zena Assaad brilliantly summarized how we got here:

1. Anthropic's Claude is the only AI model the U.S DoW currently uses for their classified systems.
2. Claude was allegedly used during the raid of the former Venezuelan president in January.
3. For months, Anthropic and the DoW have been negotiating the terms under which the military can use Claude.
4. Anthropic want to ensure their AI models are not used for mass surveillance on American citizens or for developing weapons that fire with no human involvement.
5. The DoW claim these conditions are too restrictive and the many grey areas of military operations would make it infeasible to meet these terms.
6. The DoW have deemed Anthropic a "supply chain risk". Formerly designated the company a supply chain risk would create a subcontracting nightmare, particularly for any company which uses Claude - the current AI model the DoW uses extensively.
7. Google, OpenAI and xAI have all expressed their willingness to accommodate the DoW's demands. Importantly noting, last year Google dropped their promise to not use AI for weapons or surveillance and this year OpenAI deleted the word safety from their mission.

U.S. Defense Secretary Pete Hegseth met with Anthropic CEO Dario Amodei on Tuesday. He specifically mentioned the Pentagon's claim that Anthropic raised concerns to its partner Palantir over the use of Claude during the Maduro raid. Palantir has recently been in the press for its $1B contract with the U.S. Department of Homeland Security and their tool fed on Medicare data, currently in use by I.C.E. Apparently the AI locker room gossip particularly irked Hegseth and he "gave the company until Friday to back down on its AI safeguards".

I'm not a fan of any AI company, but I've admired Anthropic's efforts in transparency. That's what made it so disappointing to see that on the same day of the meeting, Anthropic updated their Responsible Scaling Policy, saying shortcomings in the prior version could hinder its ability to compete in a rapidly growing AI market. The new version removed a stipulation that Anthropic should pause training more powerful models if their capabilities outstripped the company’s ability to control them and ensure their safety

Apparently, these changes aren't enough to put Anthropic back in the Pentagon's good graces. The Pentagon asked two major defense contractors on Wednesday to provide an assessment of their reliance on Claude — a first step toward a potential designation of Anthropic as a "supply chain risk". This is after the company expressed "willingness to work with the DoW" but insisted on blocking Claude's use for the mass surveillance of Americans or to develop weapons that fire without human involvement.

Privacy in the face of surveillance tech

If you're taking issue with my metaphor of AI and firearm safety, then consider AIs weaponization. The ACLU published documents obtained from the Department of Homeland Security shedding light on how ICE, Customs and Border Protection, and other parts of DHS have bought access to huge amounts of highly sensitive location data harvested from people’s cell phones that enables government tracking. It was not obtained through legal channels– rather bought from data brokers. It's not controversial to suggest that protestors and dissidents are being tracked as DHS demands social media platforms hand over information about anti-ICE accounts.

Your political affiliations shouldn't affect your ability to see AI being positioned as a political weapon. You should ask why politically motivated founders like Peter Thiel launch new projects like the age-verification software Persona. You should absolutely ask why a software used to confirm your age on Discord, Reddit, or Roblox needs biometric IDs.

Especially when they're sloppy doing it! Researchers investigating Discord’s age-verification checks say they discovered an exposed frontend belonging to Persona, the identity-verification vendor used by Discord. To demonstrate the privacy implications, researchers took a closer look and found a publicly exposed Persona frontend on a US government–authorized server, with 2,456 accessible files.

One LinkedIn user went through the Persona verification process, then did a deep dive into how the data was used. Here's what they collected in a 3 minute identity check:

• full name — first, middle, last

• passport photo — the full document, both sides, all data on the face of it

• selfie — a photo of their face taken in real-time

• facial geometry — biometric data extracted from both images, used to match the selfie to the passport

• NFC chip data — the digital info stored on the chip inside my passport

• national ID number

• nationality, sex, birthdate, age

• email, phone number, postal address

• IP address, device type, MAC address, browser, OS version, language

• geolocation — inferred from IP

And then there’s the weird stuff:

• Hesitation detection — they tracked whether the user paused during the process

• Copy and paste detection — they tracked whether the user was pasting information instead of typing it

Persona ran a background check on them and then shared that data with SEVENTEEN subprocessors including OpenAI, Anthropic, AWS, and Google Cloud platform. Persona also shared the data with:

• Persona’s service providers — vendors working on their behalf
• Their “global network of data partners” — those third-party databases work both ways
• Affiliates and subsidiaries — related companies that share “common data systems”
• Anyone who buys Persona — in a merger, acquisition, or bankruptcy, your data goes with the deal
• Law enforcement

Left on read deadline

Anthropic was given until Friday to agree to the Pentagon's terms of blind faith. If they refused, Secretary of Defense Pete Hegseth promised to designate the company a "supply chain risk" and cut their contract as the Pentagon's only approved model.

On Thursday, CEO Dario Amodei took to the company's blog, writing:

"Anthropic understands that the Department of War, not private companies, makes military decisions. We have never raised objections to particular military operations nor attempted to limit use of our technology in an ad hoc manner.

However, in a narrow set of cases, we believe AI can undermine, rather than defend, democratic values. Some uses are also simply outside the bounds of what today’s technology can safely and reliably do. Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now"

The AI company Antrophic looked at the Pentagon and drew a line in the sand. Claude will not be used for mass domestic surveillance or fully autonomous weapons. According to Amodeli, these use cases would undermine, rather than defend, democratic values and are outside of the bounds of what today’s technology can safely and reliably do.

This did not go over well.

The day after Anthropic's post, the U.S. President wrote on his social media platform, "I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic's technology," That announcement is particularly extraordinary because Claude is the only AI model currently used in the military's classified systems.

Secretary Hegseth then tweeted, "This week, Anthropic delivered a master class in arrogance and betrayal as well as a textbook case of how not to do business with the United States Government or the Pentagon.

Our position has never wavered and will never waver: the Department of War must have full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic….

…In conjunction with the President’s directive for the Federal Government to cease all use of Anthropic’s technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security. "

The senior Pentagon official who has been steering the negotiations with Anthropic and other AI firms, responded by calling Amodei a "liar" with a "God complex" who was "ok putting our nation's safety at risk."

Anthropic returned to their blog to pen a response apt for a dark comedy:

"We have not yet received direct communication from the Department of War or the White House on the status of our negotiations.

We have tried in good faith to reach an agreement with the Department of War, making clear that we support all lawful uses of AI for national security aside from the two narrow exceptions above. To the best of our knowledge, these exceptions have not affected a single government mission to date."

Significant efforts in social media posts, but apparently no one in the administration thought to email them back? It was certainly a fast-moving kerfuffle.

The deadline was 5:01 PM, Eastern time. At 9:56 PM ET, Sam Altman tweeted, "Tonight, we reached an agreement with the Department of War to deploy our models in their classified network." (xAI already signed an agreement to let the military use its model, Grok, in classified systems.) The following day, U.S. military used Anthropic's Claude to support Operation Epic Fury against Iran.

Open AI's play to be the 'good guy'

After Anthropic's defiant declaration, more than 200 employees from Google and OpenAI have signed an open letter stating their solidarity. Soon users began to protesting, taking to social media platforms with #quitGPT. 1.5M accounts cancelled. ChatGPT uninstalls surged by 295% after the DoD deal. News of OpenAI co-founder and president's status as a top donor to the MAGA Inc. superpac donor resurfaced.

On 2 March, Altman announced that OpenAI reworked their agreement with the Pentagon on Monday night governing the Defense Department’s use of its AI services. Their updates to the partnership announcement read:

"We have three main red lines that guide our work with the DoW, which are generally shared by several other frontier labs:

• No use of OpenAI technology for mass domestic surveillance.
• No use of OpenAI technology to direct autonomous weapons systems.
• No use of OpenAI technology for high-stakes automated decisions (e.g. systems such as “social credit”)."

I can only imagine the confused faces over at Anthropic as they read it. The FAQs on the announcement include classics like "Why are you doing this?" to which OpenAI wrote, "... We also wanted to de-escalate things between DoW and the US AI labs."

That's right. Anthropic's negotiation was left on rad and OpenAI added a bullet to the "no-go" list, but they're just jumping on a grenade for US AI. Way to take one for the team, buddy.

Claude hit No. 1 in the App Store. Multiple trade groups representing tech, software and AI companies in Washington warned Defense Secretary Pete Hegseth on Wednesday against designating U.S. companies as a supply chain risk. Amongst the signers are Apple, Adobe, Amazon, CoreWeave, Google, Meta, and Microsoft.

The Pentagon/Anthropic dispute should not be framed as patriotism versus corporate virtue. It is about whether law, safety and ethics, will meaningfully shape the responsible integration of, and limitations for, AI into warfare. The path forward needs to be responsible-by-design, anchored in law and ethics.

Don't ask me to get hype if that comes with an expectation to ignore what's actively happening. Playing dumb won't keep you safe. Regardless of political affiliations, I think we can all agree safety mechanisms are a good thing.