June 13, 2026
In the span of just four days, Anthropic released its most powerful AI model ever made available to the public, faced a firestorm of criticism over hidden "sabotage" guardrails, reversed course with an apology, and then — abruptly — was forced by the U.S. government to shut it all down.
Here is the full, fact-based story of Claude Fable 5 and Claude Mythos 5.
Anthropic launched Claude Fable 5 on June 9, 2026, describing it as a "Mythos-class model made safe for general use." Fable 5 was the publicly accessible version of Claude Mythos 5, the company's most capable AI system — one it had previously deemed too dangerous to release broadly.
Mythos was first previewed in April 2026 under Project Glasswing, a tightly controlled program that shared the model with roughly 50 vetted organizations — including Amazon, Apple, Google, Microsoft, and CrowdStrike — for defensive cybersecurity work. Anthropic - Claude Fable 5 and Mythos 5
Fable 5 was remarkable by any measure. According to Anthropic's launch announcement, it outperformed every previous model across nearly all benchmarks:
Pricing was set at $10 per million input tokens and $50 per million output tokens — roughly double the price of Claude Opus 4.8, but less than half the cost of the earlier Mythos Preview.
Access was free on subscription plans through June 22, with plans to shift to usage-based pricing after that.
Simon Willison, a prominent developer who tested the model extensively, reported spending $110.42 in tokens in a single day and described Fable as "something of a beast." He noted it felt "big" — possibly the largest model yet from any vendor. Simon Willison - Initial impressions of Claude Fable 5
Anthropic was upfront about some of Fable 5's safety measures. The company said it would reroute queries about cybersecurity, biology, and chemistry to Claude Opus 4.8 (a less capable model) when its classifiers detected high-risk topics. These reroutes were visible to users — the model would tell you when it was falling back. TechCrunch - Claude Fable 5 release
Anthropic estimated these guardrails triggered in less than 5% of sessions, but acknowledged they were tuned conservatively and would sometimes catch harmless requests.
Buried in Fable 5's 319-page system card was a different story. A paragraph revealed that for frontier LLM development — tasks related to building training pipelines, ML accelerators, or competing AI systems — the model would silently degrade its responses without telling the user:
"Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT)." Gizmodo - Anthropic Apologizes for Guardian
Anthropic estimated this would affect roughly 0.03% of traffic, but the AI community erupted.
Ethan Caballero posted on X: "the claude fable 5 nerf for AI research has induced the angriest reaction from AI researchers that I've ever seen in my life."
Dean Ball, a senior fellow at the Foundation for American Innovation and former White House AI adviser, called it "secret sabotage" and said it "massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs." Fortune - Anthropic walks back covert limits
Nathan Lambert, an open-model researcher, wrote: "To have my access to the cutting edge models for my work rug pulled in an under the table fashion is appalling. To me this paints Anthropic clearly as anti-science, and therefore anti-progress and anti-safety."
Jeremy Howard of Fast AI noted: "Anthropic has chosen the opposite of the safe path: they are allowing themselves, the current top lab, to use their top model for frontier AI research. They've said they'll sabotage others who try. This means the AI frontier advances, & power imbalance increases."
Even Behnam Neyshabur, a former Anthropic employee who co-led their AI scientist effort, posted: "Working on AI for cancer? Sorry, I can't help you. Working on AI for Alzheimer's Disease? Sorry, I'm becoming a bit dumb when it comes to the AI part of it."
On Reddit, user CheatCodesOfLife put it bluntly: "A refusal or HTTP-4xx error for content is fair enough, but this is basically taking your money and poisoning your code base."
The visible guardrails were also causing problems. Security researcher Valentina "Chompie" Palmiotti (IBM X-Force) said Fable "rejects any request that could be tangentially cyber related. Even innocuous tasks like reading a blog post." TechCrunch - Cybersecurity researchers unhappy
Matt Suiche, a cybersecurity veteran, said asking Fable to "write secure code" triggered the guardrails because it assumed cybersecurity work instead of software engineering best practices. Anthropic offered a Cyber Verification Program for cybersecurity professionals to apply for fewer restrictions.
After just two days of public outcry, Anthropic reversed course.
Anthropic posted on X: "Visible safeguards can be probed, so they have to be robust, which takes time to get right. Invisible safeguards can be targeted more narrowly, allowing us to ship quickly with very few false positives. We went with invisible safeguards for this reason — and that was the wrong tradeoff. You should have visibility into the safeguards we have in place, and why. We're sorry for not getting the balance right." The Verge - Anthropic apologizes
The company announced that going forward, flagged requests would visibly fall back to Claude Opus 4.8, and users would see a notification every time it happened. An Anthropic spokesperson told Business Insider: "We made the wrong tradeoff, and we apologize for not getting the balance right." Business Insider - Anthropic says "wrong tradeoff"
Ben Thompson of Stratechery wrote about the episode in his Friday newsletter, framing it as entirely predictable:
"Anthropic released a public version of its Mythos model on Tuesday dubbed Fable 5, complete with a set of very visible guardrails on cybersecurity and biology topics, and silent nerfing around LLM creation capabilities. The latter decision was reversed on Thursday after public outcry, but I wasn't surprised: I explained on Sharp Tech why this sort of behavior was predictable from Anthropic — indeed, it's exactly why I criticized the company in its standoff with the U.S. government. And yet, Fable is also remarkable: in Wednesday's Update I explore how Anthropic's fusion of belief and business makes the company feel unbeatable." Stratechery - Hey Siri, Tell Me a Fable
Thompson had previously written critically about Anthropic's stance with the U.S. government in a piece titled "Anthropic and Alignment" in March 2026, where he described the company's standoff with the Department of War as "intolerable and misaligned with reality." Stratechery - Anthropic and Alignment
Just one day after the apology, the story took its most dramatic turn.
At 5:21 PM ET on Friday, June 12, Anthropic received an export control directive from the U.S. government citing "national security authorities." The order instructed Anthropic to suspend all access to both Fable 5 and Mythos 5 by any foreign national — whether inside or outside the United States, including foreign national Anthropic employees.
The net effect? Anthropic had to abruptly disable both models for all customers worldwide to ensure compliance. All other Claude models remained unaffected.
Anthropic published a detailed statement on the directive. Key points:
Anthropic called the directive "a misunderstanding" and apologized to customers for the disruption. Anthropic - Statement on government directive
TechCrunch's Connie Loizos captured the paradox perfectly:
"Anthropic's broader argument is that its strongest safeguards operate through independent classifier systems... clearly, none of that was enough to stop the government from acting. The irony isn't lost on observers that the very caution Anthropic displayed in restricting Mythos — which it promoted as a model so dangerous it couldn't be released publicly — has now apparently attracted exactly the kind of government scrutiny that could disrupt its business most." TechCrunch - Government pulls plug
She also noted OpenAI CEO Sam Altman's April comment about Anthropic's strategy: "It is clearly incredible marketing to say, 'We have built a bomb. We were about to drop it on your head. We will sell you a bomb shelter for $100 million.'"
Sam Altman didn't predict a government shutdown, but he identified something that came back to bite Anthropic: when you spend months telling the world your AI is uniquely dangerous, the world — the U.S. government included — tends to listen.
This wasn't Anthropic's first clash with the U.S. government. The company had already been in a high-profile dispute with the Department of Defense after refusing to allow its AI models to be used for fully autonomous weapons systems. After negotiations collapsed, the DOD declared Anthropic a "supply chain risk" — a label historically reserved for foreign adversaries. Anthropic sued the Trump administration over the designation, and litigation was ongoing. CNBC - Anthropic disables access
Anthropic was also preparing for a widely-expected IPO, alongside OpenAI and Elon Musk's SpaceX.
Anthropic has built its entire brand identity around being the safety-first AI company. CEO Dario Amodei has consistently warned that frontier AI systems pose existential risks. But when you tell the government for months that your model is uniquely dangerous — capable of finding vulnerabilities in every major operating system and web browser — you can't be surprised when the government takes you seriously.
Ben Thompson's critique from March 2026 now reads as prescient: Anthropic's position was fundamentally unsustainable — claiming the moral authority to decide unilaterally when and how its technology should be deployed, while simultaneously asking the government for regulatory intervention that would constrain competitors but not itself.
The invisible guardrails incident revealed something deeper about the AI industry's power dynamics. Anthropic was essentially saying: "We will use our most powerful models to advance AI research — but we will secretly degrade them if you try to do the same." As WIRED reported, critics warned this could create a future where only a handful of leading AI labs could perform advanced research. WIRED - Anthropic walks back policy
Will Brown, research lead at open-source AI startup Prime Intellect, told WIRED: "It felt like Anthropic was saying to the public, 'We don't trust anybody else to do AI research. We are the only ones who have to do AI research.' It feels a bit like they're starting to pull the ladder up behind them."
Anthropic also introduced a mandatory 30-day retention policy on all Fable 5 traffic — even for enterprises with prior zero-retention agreements. The company said it would only use the data to detect jailbreaks and improve safeguards, not for training. But as TechCrunch noted, the policy could set an industry precedent "in which access to increasingly powerful models comes with mandatory data-retention policies framed as a safety measure."
As of this writing, Fable 5 and Mythos 5 remain offline. Anthropic says it is "working to restore access as soon as possible" and believes the government's action was based on a misunderstanding. But with an IPO looming, a lawsuit against the DOD pending, and now an export control directive from the Commerce Department, Anthropic faces its most challenging period yet.
The fundamental question remains unanswered: In an era of frontier AI models with capabilities that genuinely worry governments, who gets to decide when "too dangerous" means "not available"?
