For eighteen days, the most capable AI model on the market sat behind a government lock.That lock came off this week. On 30 June, the US Commerce Department lifted the export controls it had imposed on Anthropic's Claude Fable 5 and its restricted sibling Claude Mythos 5, and by Wednesday afternoon Fable 5 was live again for users worldwide — VentureBeat clocked the official announcement at 3:31 pm Eastern on 1 July. Commerce Secretary Howard Lutnick framed the resolution on X: "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI."The return, though, is a story with fine print. The model comes back wearing a new safety classifier, Anthropic comes back carrying a set of formal commitments to Washington, and paying subscribers come back to a countdown: included access to Fable 5 ends on 7 July, after which the model runs on paid usage credits — a squeeze that has the Claude subscriber base openly furious. And the episode leaves a mark bigger than any product: for close to three weeks, the world watched a government switch off a frontier AI model by decree, then switch it back on under conditions. What follows is the whole picture — the shutdown, the fix, the money, and the precedent.Key TakeawaysThe US Commerce Department lifted export controls on Claude Fable 5 and Claude Mythos 5 on 30 June, and Fable 5 returned to users worldwide on 1 July across Claude. ai, the Claude Platform, Claude Code and Claude Cowork.The models had been offline since 12 June, when an export directive banned access for any foreign national — and Anthropic, lacking a real-time way to verify nationality, switched both off for everyone.The trigger was a report in which Amazon researchers found a method of bypassing Fable 5's safeguards to identify software vulnerabilities; Anthropic's counter is that every model it tested, including rivals, could produce the same demonstration.Fable 5 comes back with a new safety classifier Anthropic says blocks the reported technique in more than 99 per cent of attempts, at the cost of more false alarms on routine coding work.The included window is short: Pro, Max, Team and select Enterprise plans get Fable 5 for up to 50 per cent of weekly usage limits through 7 July, after which access runs on paid usage credits — and subscribers are openly angry about it.Mythos 5, the same model with fewer guardrails, is restored only for approved US organisations, with reports of select Indian agencies and firms gaining access under Project Glasswing.What Actually Happened?The sequence, assembled from Anthropic's own accounts and the reporting around them, runs like a thriller compressed into 22 days.Anthropic launched Fable 5 and Mythos 5 on 9 June. The two share one underlying model: Fable 5 shipped with strong safeguards for general use — the company's first release of a model this capable to the public — while Mythos 5, with fewer guardrails and what Anthropic called the strongest cybersecurity capabilities of any model in the world, went only to trusted partners in Project Glasswing, the company's defensive-security programme run with the US government. The early customer stories were extraordinary: Stripe reported that Fable 5 compressed a migration across a 50-million-line Ruby codebase into a single day, work Anthropic said would have taken a team over two months by hand.Three days later, the lights went out. At 5:21 pm Eastern on 12 June, per VentureBeat's reconstruction, the US government issued an export-control directive citing national security authorities, banning access to both models by any foreign national — inside or outside American borders, a sweep that covered Anthropic's own non-citizen staff. Anthropic lacked a real-time way to verify a user's nationality at the API layer, so compliance meant one thing: it switched both models off for everyone, everywhere.The trigger, per Anthropic's 30 June account, was a report in which Amazon researchers had found a method of bypassing Fable 5's safeguards — prompting it to identify a number of software vulnerabilities, and in one case producing code demonstrating how a vulnerability could be exploited. The Wall Street Journal reported the Amazon findings, and 9to5Mac, citing that reporting, noted that conversations between Amazon chief executive Andy Jassy and the White House reportedly helped prompt the directive. The subplot writes itself: Amazon is one of Anthropic's largest investors and a Glasswing partner, and per that reporting its researchers' work put Anthropic's flagship on ice. Treat the Jassy detail as reported rather than confirmed — but the irony has been widely noted.WhenWhat happened2 JunUS executive order creates a voluntary pre-release review path for frontier AI models9 JunAnthropic launches Claude Fable 5 (general use) and Mythos 5 (restricted, Project Glasswing)12 JunExport-control directive bans foreign-national access; Anthropic takes both models offline globally26 JunCommerce approves restored Mythos 5 access for select US organisations30 JunExport controls lifted; Anthropic publishes its redeployment plan1 JulFable 5 returns worldwide on Claude. ai, Claude Platform, Claude Code and Claude Cowork7 JulIncluded Fable 5 access ends for subscription plans; usage credits take overThe Jailbreak, And Anthropic's CounterpunchUnderstand the technical dispute and the politics make sense. A jailbreak, in this context, is a prompting technique that gets a model to do something its safety training is meant to head off. The Amazon report showed Fable 5 could be steered into vulnerability-hunting, and, once, into demonstrating an exploit — exactly the offensive-cyber territory that had kept Mythos-class capability behind Glasswing's velvet rope.Anthropic's defence was pointed. Its own testing, the company said, confirmed that many less capable models — including Claude Opus 4.8, GPT-5.5 and Kimi K2. 7 — could identify the same vulnerabilities as Fable 5 did in the report. On the single exploit demonstration, Anthropic went further: every model it tested could produce the same demonstration, a list that ran from its own Haiku 4.5 and Sonnet 4.6 through OpenAI's GPT-5.4 and GPT-5.5 to China's Kimi K2. 7. In the company's telling, the flagged behaviour was routine defensive-security work available across the industry, rather than a hidden super-capability unique to Fable.The fix that broke the logjam was engineering rather than argument. Anthropic said it "moved quickly to address the reported bypass," deploying what it described as an "improved safety classifier that targets and blocks the behavior described in the report" — a classifier being an automated filter that inspects a request and decides whether it is harmless or an attempt to slip the guardrails. Per Anthropic's 30 June write-up, the new classifier stops the reported technique in more than 99 per cent of attempts; blocked requests fall back to the less capable Opus 4.8, with the user told. The company itself flags the trade-off: benign requests may get caught during routine coding and debugging, with false positives it promises to grind down. Users had grumbled even before the ban that Fable was quick to swerve away from cybersecurity and biology questions; the new filter sharpens that tension rather than dissolving it.Government sign-off came in two stages. On 26 June, Lutnick wrote to Anthropic — in a letter viewed by CNBC — that "appropriate safeguards" were in place to let certain "trusted partners" use Mythos 5 again. On 30 June, Commerce lifted the controls outright, and Anthropic announced the redeployment on X: "After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks."The Meter Starts On 7 JulyFor the everyday Claude subscriber, the return reads like power restored after a storm — with a new meter bolted to the wall and the dial spinning faster than before.The restoration itself is broad. Fable 5 is live worldwide on Claude. ai, the Claude Platform, Claude Code and Claude Cowork, with Anthropic promising to re-enable the model on AWS, Google Cloud and Microsoft Foundry as quickly as possible. But the current flows at half strength, and on a timer. For Pro, Max, Team and select Enterprise plans, Fable 5 is included for up to 50 per cent of weekly usage limits through 7 July. After that, the included supply is cut and access runs on usage credits — metered, pay-as-you-burn power. Pricing trackers put the Fable 5 rate at 10 dollars per million input tokens and 50 dollars per million output tokens, double Opus 4.8's rate; verify those figures against Anthropic's published rate card, since the company has yet to publish what "up to 50 per cent of weekly usage limits" converts to in tokens or money for any given plan. Enterprise customers should read their wiring diagram with care: premium Enterprise seats share the grace window, while standard Enterprise seats sit outside it entirely — their access runs on usage credits from day one, and Fable 5 stays dark for organisations that leave credits switched off.SurfaceStatusClaude. ai, Claude Platform, Claude Code, Claude CoworkLive worldwide from 1 JulyAWS, Google Cloud, Microsoft FoundryRe-enabling "as quickly as possible"Subscription plans (Pro, Max, Team, select Enterprise)Included up to 50 per cent of weekly limits through 7 JulyAfter 7 JulyPaid usage creditsMythos 5Approved US organisations since 26 June; Glasswing expansion under wayThe subscriber base has done the arithmetic, and the mood is sulphurous. When Fable 5 launched on 9 June, the original plan gave subscribers a two-week included window with a full allowance. The ban swallowed most of it — users got roughly three days — and the model has returned with a shorter window and a new 50 per cent cap stacked on top. PCWorld, surveying the backlash, quoted the Reddit thread where it boiled over: "Not a good look to bring Fable back and then both half the usage and take away days," wrote one user; "We got to use it for like 3 days out of the 14 we were told, and now we get it for just 7 days at half usage?" asked another. Anthropic, for its part, closed its redeployment post by saying it is "grateful to our users for bearing with us through this disruption." Gratitude, the thread suggests, was priced in dollars the users expected to keep.A Licence With ConditionsLook past the billing and the deal that freed Fable 5 resembles a provisional driving licence: the model is back on the road, but the examiner kept the paperwork, wrote conditions on it, and reserved the right to call the driver back in.The conditions are substantial. Per VentureBeat's summary of the commitments, Anthropic will work with the US government on protocols, standards and releases for Mythos, Fable and future models; inform the government of malicious activity; expand pre-release government access and evaluation for frontier models; share information rapidly when significant jailbreaks or misuse surface; fund joint research; and work toward a common industry security bar. The Commerce Department, in turn, explicitly reserved the right to re-evaluate the permissions and re-impose licence requirements if circumstances change. Anthropic has also opened a HackerOne programme paying researchers to report fresh Fable 5 jailbreaks, and has begun drafting — alongside Amazon, Microsoft, Google and other Glasswing partners — a consensus framework for grading the severity of AI jailbreaks and how developers should respond, rules it hopes will be codified in strong regulation applied equally across frontier labs, and even serve, in the company's words, as a template for global coordination.The road everyone now drives on was paved in early June. A 2 June executive order created a voluntary path for companies to submit frontier models for government review before release, giving agencies 60 days to build the machinery — and the Fable episode became its first live-fire test. The pattern is spreading: days earlier, OpenAI previewed GPT-5.6 to a small, government-approved group rather than the public, citing the same dual-use worry. And the whole affair unfolded against a competitive clock — the crackdown coincided with the swift rise of Chinese open-source models that are nearly as capable and far cheaper, and a number of tech executives and investors argued the pause gifted Chinese developers valuable catch-up time. Eighteen days is an eternity in a race measured in weeks.India's Seat At The TableFor Indian readers this is local news wearing a Washington byline, because India is Anthropic's second-largest market — and its enterprises spent 18 days learning what a single foreign directive can do to their toolchain.The corporate stakes are concrete. Anthropic recently signed a deal with TCS to equip 50,000 employees with its models, and separately announced a collaboration with Infosys to deploy enterprise AI built on the Claude family — both bets that presume the models stay switched on. More striking is the Glasswing thread: ETV Bharat, citing sources familiar with the matter, reports that select Indian government agencies and private firms were recently granted access to the Mythos model under the programme — which, if borne out, would place Indian entities inside the most restricted tier of American AI access at the exact moment Washington is formalising who gets to hold the keys. Treat that as sourced reporting awaiting official confirmation from either government.The 7 July cliff lands here too. Indian Pro and Max subscribers face the same halved allowance and the same switch to usage credits, priced in dollars — and with Anthropic yet to publish the credit conversions, Indian teams budgeting for Fable 5 are planning against unpublished numbers. For the CIO who just promised the board a Fable-powered quarter, the past three weeks carried a plainer lesson: the most capable model in the world can vanish from the stack overnight, on the strength of a letter from a ministry ten time zones away.When Launches Become NegotiationsStep back from the billing dates and the signed conditions, and the durable story is the precedent.VentureBeat's post-mortem lands on the structural point: the two-week shutdown exposed how fragile centralised, closed-API models are as business infrastructure, with enterprise pipelines hostage to sudden regulatory shifts — and frontier launches starting to look less like product releases than negotiated deployments shaped by national-security review. Every element of this week's resolution confirms the shift. The model returned with a government-audited filter. The company returned with signed commitments and a standing invitation for pre-release inspection. The regulator kept a documented right to pull the permission back. And the industry's response — a shared jailbreak-severity framework drafted by rivals — treats the next incident as a certainty to be managed rather than a possibility to be argued about.The date on every subscriber's calendar is 7 July, when the included window closes and the billing begins. But the date historians will circle is 12 June — the evening a government proved, in 22 minutes of directive, that the most advanced artificial intelligence on Earth ships at the state's pleasure. Fable 5 is back. The world where it could leave and return on a company's sole authority is what stayed switched off.FAQWhat are Claude Fable 5 and Claude Mythos 5?They share one underlying model — Anthropic's most capable to date. Fable 5 ships with strong safeguards for general use, routing some sensitive queries to the less capable Opus 4.8. Mythos 5 carries fewer guardrails and goes only to approved organisations, mainly for defensive cybersecurity under Project Glasswing.Why were the models taken offline?A 12 June US export-control directive banned access by any foreign national after the government learnt of a report in which Amazon researchers bypassed Fable 5's safeguards to identify software vulnerabilities. With zero real-time means of verifying user nationality, Anthropic switched both models off for everyone.What was the Amazon jailbreak?A prompting technique that steered Fable 5 into identifying software vulnerabilities and, in one case, producing code demonstrating how one could be exploited. Anthropic counters that every model it tested — including its own smaller models, GPT-5.5 and Kimi K2. 7 — could produce the same demonstration.What changed to bring Fable 5 back?Anthropic deployed a new safety classifier it says blocks the reported technique in over 99 per cent of attempts, agreed to a set of commitments with the US government, and opened a bug-bounty programme for new jailbreaks. Commerce lifted the controls on 30 June, and the model returned worldwide on 1 July.What happens on 7 July?Included access ends. Through 7 July, Pro, Max, Team and select Enterprise plans get Fable 5 for up to 50 per cent of weekly usage limits; after that, the model runs on paid usage credits. Standard Enterprise seats sit outside the included window altogether and need credits enabled from the start.How much will Fable 5 cost after 7 July?Pricing trackers cite 10 dollars per million input tokens and 50 dollars per million output tokens — double Opus 4.8. Anthropic has yet to publish what the in-app allowance or credit conversion equals per plan, so verify against its official rate card before budgeting.Is Mythos 5 available in India?ETV Bharat, citing sources, reports select Indian government agencies and private firms were recently granted Mythos access under Project Glasswing. Official confirmation is pending, and general availability stays off the table — Mythos remains restricted to approved organisations.Could the US government switch the models off again?Yes. The Commerce Department explicitly reserved the right to re-evaluate its permissions and re-impose licence requirements if circumstances change or commitments slip — which is why this episode reads as a template, rather than a one-off.end of article