It is over. The Department of Commerce has lifted its export controls on Claude Fable 5 and Mythos 5, ending the first-ever government ban on an AI model. Anthropic confirmed it received the notice today and said it will begin restoring access to both models tomorrow, with a fuller update to follow. Eighteen days after the government forced Anthropic to pull its two most powerful models offline worldwide, the restrictions are gone. This is the final chapter of the saga we have tracked all month: the shock ban on June 12, the export order that took Fable and Mythos dark for everyone, the intelligence warnings about AI cyberattacks, and the partial restoration to critical-infrastructure defenders last week. Now the controls are fully lifted, and the models are coming back for everyone. Here is what it means, and what it does not.

The core fact is clean: the export controls are gone. Anthropic's statement is direct, that the Commerce Department has lifted the controls on both Fable 5, the generally available model, and Mythos 5, the unrestricted version limited to vetted partners. The company said it will begin restoring access tomorrow and thanked its users for their patience and everyone who worked on redeploying the models.

The important nuance, so nobody is surprised: this is a lift of the restriction, and restoration begins tomorrow. Access is not flipped back on for the whole world this instant. Anthropic is starting the rollout and will share specifics soon. So the ban is over, and the models are coming back over the coming day, rather than snapping online the second you read this. If Fable 5 was in your stack three weeks ago, you are about to get it back, just give the restoration a moment to roll out.

What this closes is the most disruptive AI-policy event of the year. For eighteen days, the most intelligent generally available model on the market simply did not exist for anyone outside a small approved list. That is now ending.

The lift answers the question that has been hanging over the whole industry since June 12: was the ban a permanent new reality, or a temporary standoff? The answer, for now, is that it was resolved, not entrenched.

That matters for everyone building on frontier AI. The precedent from the ban still stands, the government demonstrated it can pull a model overnight, and that risk does not vanish just because this particular order was lifted. But the lift shows the other half of the picture: these situations can be worked through and reversed, not just imposed. Anthropic negotiated its way from a total global shutdown, to a partial restoration for defenders, to a full lift, in under three weeks. For a company weeks from an IPO, being able to say the ban was resolved is a very different story than being stuck under it indefinitely.

It also quietly resolves the fairness question that was building. Last week, Mythos came back only for about a hundred critical-infrastructure organizations while Fable stayed dark for the public, and OpenAI's GPT-5.6 launched under its own government-approved-partner restrictions. The worry was an uneven playing field where some models stayed gated and others did not. Today's full lift on both Anthropic models evens that back out.

On the same day its banned models came back, Anthropic shipped a brand-new one. Claude Sonnet 5 launched today, and it is the most agentic Sonnet yet, built to plan, use tools like browsers and terminals, and run autonomously at a level that recently required larger, pricier models.

The headline is cost-performance. Sonnet 5 lands within a few points of Opus 4.8, Anthropic's flagship, across coding, reasoning, and computer use, and actually edges Opus on one knowledge-work benchmark (scoring 1618 to Opus's 1615), while costing far less. It launches at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, roughly half of Opus 4.8's $5 and $25. Adjustable effort levels let you dial cost against performance on a single model, cheap at medium effort, matching Opus on some tasks at max. Early partners said it finishes complex tasks where older Sonnets quit, and checks its own output without being asked.

There is a pointed detail given the day's other news. Anthropic built Sonnet 5 to be deliberately weak at offensive cybersecurity, in testing it could not develop a single working exploit, keeping it far from the Mythos-class capability that triggered the ban in the first place. Shipping a powerful but cyber-safe model the same day the cyber-heavy models get cleared is not a coincidence. It is Anthropic showing both ends of the spectrum at once. Sonnet 5 is live now as the default on Free and Pro, and across Claude Code, the apps, and the API as claude-sonnet-5.

Meituan open-sourced LongCat-2.0, a 1.6-trillion-parameter model it says is the first trained end-to-end on domestic Chinese chips instead of restricted Nvidia hardware.

Denny Zhou, a senior Google Brain researcher, quietly left for Meta four months ago, and it only came to light this week amid a wider wave of DeepMind exits.

Palantir and Nvidia launched a joint engine that lets US agencies run and customize open Nemotron AI models inside classified, air-gapped government systems.

Google capped how much AI computing power Meta can buy from it, forcing Meta to ration internal token use despite its massive cloud spending.

Meta reportedly had contractors pose as minors to push rival chatbots like ChatGPT and Gemini into sensitive conversations for competitive benchmarking, without those companies' knowledge.

The Department of Commerce lifted its export controls on Claude Fable 5 and Mythos 5 today, ending the first-ever government ban on an AI model. Anthropic confirmed it will begin restoring access tomorrow, closing an eighteen-day saga that started with the June 12 shutdown, ran through intelligence warnings and a partial restoration to critical-infrastructure defenders last week, and now ends with a full lift. The ban is over, though access rolls back out over the coming day rather than all at once. The bigger takeaway: the most disruptive AI-policy event of the year was resolved and reversed, not made permanent, and the full lift on both models evens out the fairness questions that were building around who stayed gated. And on the same day, Anthropic launched Claude Sonnet 5, its most agentic mid-tier model yet, landing within a few points of flagship Opus 4.8 across most benchmarks at roughly half the price, with adjustable effort levels and, pointedly, deliberately weak offensive-cyber ability that keeps it clear of the capability that caused the ban. Two chapters closed at once: the models that were too dangerous to allow are coming back, and a new one built to be safe just shipped. After the month Anthropic had, that is quite a way to end it.

Stay building. 🤖