CONVERGER #4.5: The Anthropic Fable FUBAR
How the White House is Censoring Our AI Future
Welcome to CONVERGER, a biweekly-ish newsletter mapping the content singularity where AI and the internet collapse all media into one—a connective node where emerging technology, policy, culture, futures thinking and storytelling intersect.
<<Achievement Unlocked: CONVERGER was cited in the New York Times!>>
CONVERGER presents news and views from an AI, internet and media policy expert who is pro-innovation but anti-hype, allergic to both AI panic and AI boosterism, and passionate about supporting rather than supplanting human creativity with new technology.
This week we interrupt our regularly scheduled programming for a short pre-July 4th feature focused on the current live controversy around the White House blocking the release of frontier AI models. Read it and then go spend the weekend celebrating our freedoms while we still have them!
Next week we’ll have a full issue #5 of CONVERGER with an extensive interview on how to copyright your AI-generated content with the lawyer who did it for the AI-animated film Critterz, in-depth analysis of the new A24/Google and Lionsgate/Runway deals, a feature on Hollywood’s current buying spree for viral short stories, and more.
I’m Kevin Bankston, your host. You can watch me develop newsletter content in real-time on LinkedIn and the social network formerly known as Twitter, and less often on Bluesky and Instagram. You can also look for my more wonkish takes on AI governance at Elicitation, the new Substack from my AI policy day-job colleague Miranda Bogen of the Center for Democracy & Technology’s AI Governance Lab. (Note that my Substack articles don’t necessarily reflect CDT’s positions.)
So let’s get to it. And please share with your friends and colleagues if you enjoy!
The Anthropic Fable FUBAR: How the White House is Censoring Our AI Future
Converger isn’t a cybersecurity or digital civil liberties newsletter, and I try to keep things light and focused on the intersection of technology, media, and culture. But when the Executive Branch unilaterally targets the most powerful tool for communicating information and expression of our time—perhaps of all time—with a standardless censorship regime, I figure it’s worth talking about.
And that’s exactly what has happened—what is still happening, and escalating—which is why I organized a press conference last Tuesday, June 23, in my day-job capacity as Senior Advisor on AI Governance at the Center for Democracy & Technology.
Billed as a “Cross-Spectrum Civil Society Briefing on Trump Admin Blocking Anthropic Models,” it brought together over a half-dozen experts from across the political spectrum. Those experts from both left and right disagree on a whole lot of things, but pretty much all agree that the government’s restrictions on the release of Anthropic’s latest AI models are probably unlawful and unconstitutional and definitely dumb and dangerous.
My warning last Tuesday, that this situation would only get worse the longer it dragged on, was immediately borne out Thursday afternoon when OpenAI “voluntarily” agreed with the White House to stagger the release of its newest model, GPT 5.6, and only share it with customers that are individually approved by the government (!).
Then, perhaps to avoid charges of favoritism, on Friday afternoon the administration agreed to let Anthropic’s Mythos out of export jail, though it’s still on parole. As with OpenAI, access is restricted to whomever the White House approves: as of now, only one hundred or so U.S. companies and governmental entities will be allowed to use it. Yet there is no indication of how this decision was made, what if anything has changed in regard to the purported cybersecurity risk, why and under what standards these entities were chosen, or whether and when Fable will be available to consumers again. So we’re still in a situation where an unchecked government known for cronyism is leveraging a secret process to decide who, if anyone, can access America’s most powerful AI tools.
Meanwhile, as our own government is keeping US models out of the hands of not just foreign nationals but American citizens based on a secret and standardless process, prompting everyone on Earth who relies on US AI companies to reconsider their AI strategy, a Chinese lab just released an open source model that rivals our best proprietary models. The very models that might best help us identify and patch vulnerabilities before AI-enabled attackers exploit them are the very models being suppressed, ultimately making us less safe, as a long list of cybersecurity experts have pointed out.
This is not how we win the AI future. Nor is it how we defend what is quickly becoming everyone’s primary information portal against an overreaching government trying to dictate what lawful information we can or cannot access. When China is the nation racing ahead releasing open source frontier AI to the world, while a corruption-prone US administration with a demonstrated disregard for the rule of law is the one restricting releases by leveraging fear of political reprisal, something has fundamentally shifted in the world. And it’s not a shift that anyone who cares about free expression, democratic values, American competitiveness, open science, or shared technological progress should feel good about.
Below are my comments that opened last week’s press conference, which are equally if not more relevant this week; you can also see some choice quotes from my co-panelists here.
“On June 12, eleven [now nineteen] days ago, Anthropic was forced by the US government to pull its frontier models Fable 5 and Mythos 5. Specifically, it was required to comply with a Commerce Department directive, purportedly based on statutory export control authorities and citing national security concerns, to bar all foreign nationals—including those in the US and including Anthropic’s own employees—from accessing those models. This demand—which practically would require the models to be taken down altogether—was first made by a phone call, reportedly giving Anthropic only 90 minutes to act, and followed by a letter from Commerce Secretary Lutnick.
“In the days since, it has become clear that this decision by the government was made hastily, based on unclear evidence of a cybersecurity threat, with little to no due process, under dubious statutory authority, and in a way that raises profound constitutional questions about free expression and access to information in the AI age.
“Even worse, it was done without any clear plan—for how long this censorship would last, for what Anthropic would need to do to be allowed to put its models back up, for what to do when other models from other companies, both proprietary and open source, offer similar capabilities around identifying cybersecurity vulnerabilities—something that arguably has already happened, including with the release of ChatGPT 5.5-cyber [and now ChatGPT 5.6, too!].
“This action against Anthropic is occurring just months after Anthropic was designated as a supply chain risk for refusing to renegotiate use limits in its contract with the Department of Defense in a politically charged debate where the administration repeatedly attacked Anthropic as “woke AI.” And it occurs just weeks after the White House unveiled an AI strategy that was built around what were emphatically intended to be voluntary safety evaluations by the government. These facts raise the specter that the government simply does not have a coherent strategy, or that its choices are being driven by factors beyond just cybersecurity.
“Put simply, if the government can demand the depublishing of AI models in this manner—AI models which are becoming a primary means by which US citizens obtain information and expression—our most profound constitutional values are at risk.
“It is sadly ironic we are having this conversation today, when tomorrow is the 55th anniversary of the White House petitioning the Supreme Court to block the publication of the Pentagon Papers. In that very similar case, an overreaching White House attempted to censor the publication of newsworthy information based on vague national security concerns, a prior restraint that the Supreme Court found to be unconstitutional. The prior restraint on Anthropic’s publication of its most useful model and by extension all the information it could provide to the public is just as momentous and dangerous as that historic case of censorship, if not more so.
“The momentous nature of this event is highlighted by the range of voices we’ve brought together today to speak out against it. Our speakers come from across the civil society spectrum—from tech left to innovation right, from civil liberties to cybersecurity, from public interest to business interest—and will address the broad range of legal, policy, technical and economic concerns with the government’s actions in this unprecedented case. We don’t all agree on everything, even in regards to this case—but we do all agree that this moment is profoundly troubling, and becomes more troubling every day it drags on without a clear and lawful resolution.”
That’s my two cents, but if you want more, here are some other useful takes from other Substackers:
Neil Chilson and Adam Thierer on “What’s Worse Than an ‘FDA for AI’? (Whatever Just Happened to Anthropic).”
Dean Ball on “What Should Be Done” now that the White House has begun instituting “a de facto involuntary licensing/preapproval regime for frontier models.”
Tyler Tone of the Foundation for Individual Rights and Expression (FIRE) on the question “How does the First Amendment apply to AI?” Answer: in a lot of ways!
Nathan Lambert says “Welcome to the AGI era of AI governance”: ”It’s a one way door and we weren’t ready for it.”
Zvi Mowshowitz on how the “White House Will Ad Hoc Decide Who Can Individually Access GPT-5.6” and how that’s “a maximally terrible policy.”
Sharon Goldman on “The real fallout from the Trump administration’s AI access ‘kill switch,’ according to legal and policy experts,” including me.
LATE-BREAKING UPDATE: The night before this piece was going to press, Anthropic announced that it was told by the Commerce Department that the export restrictions on Fable 5 and Mythos 5 were being lifted. But we still don’t really know why they were blocked in the first place, how the decision was made to block or to unblock, what exactly Anthropic had to do in return, or anything else about this opaque and ad hoc process that the White House just made up to control who is allowed to access these powerful information utilities. So, all of the concerns and critiques above still stand, especially now that ChatGPT 5.6 has been restricted under the same non-process process.
What’s happens next? No one knows, except maybe Commerce Secretary Lutnick, but he’s not saying.
And that’s what’s converging this week! See you next week with many more thousands of words on the nexus between AI, copyright, creativity, and culture.