TL;DR: Just in time for the Fourth of July, Anthropic’s Fable 5 and Mythos 5 models have declared independence (sort of). The US government’s export ban—which was implemented three weeks ago—has finally been lifted. The reprieve lands right as a Chinese model matched a top US model on key cybersecurity benchmarks.

What happened: Just weeks after it was abruptly pulled, Anthropic announced that Fable will roll out again for users worldwide starting today. The government export ban came after Amazon researchers flagged a jailbreak the Trump administration claimed could threaten national security.

If you’re wondering what the ban actually achieved (and why it was able to be reversed so quickly), you’re not alone. Anthropic (unsurprisingly) disagreed with the government’s initial move, and some cybersecurity experts say that the jailbreak issue was “significantly overblown.” According to the Claude maker, other models—including Opus 4.8 and GPT-5.5—identified the same vulnerabilities Fable did during its own tests.

Second time’s the charm?: Still, don’t expect the exact same Fable you may have tested for all of three days before the suspension. Anthropic says it’s now blocked the jailbreak method (which will also block more cybersecurity tasks in general), and that “some routine tasks like coding and debugging will fall back to Opus 4.8.”

If you’re on a paid consumer plan (and some enterprise ones), you can use Fable 5 for up to half your weekly usage limit until July 7. After that, Fable will be billed pay as you go (at double the rate of Opus 4.8).

Home-field disadvantage: The reversal also arrives as the AI competition with China heats up another notch, with a Chinese model now reportedly performing on par with Mythos on certain cybersecurity tasks (like finding bugs), per the Wall Street Journal—and at a fraction of what Opus 4.8 costs.

The Trump admin is now exercising tighter control over new model releases, but critics argue this could drive companies to use less restricted Chinese models—with former Facebook Chief Security Officer Alex Stamos calling the Mythos-Fable suspension an “own goal” for the US.

Bottom line: Anthropic has likely been on the edge of its seat waiting for the government’s green light on its priciest flagship model—especially as it races to turn a profit ahead of its IPO, and the AI competition intensifies both at home and abroad. —WK

Also at Anthropic…

• If you’re on a tight AI budget, the company just announced a discount on the newly launched update to its cheaper Sonnet model.

• Astronomers are using surprisingly lightweight AI to find new galaxies. Instead of general-purpose LLMs, astronomers are now developing and using specialized AI tools that require little computing power—and it’s happening across the field. The European Space Agency ran a neural net over 36 years of Hubble Space Telescope archives, surfacing hundreds of new objects (including previously unseen galaxies) on a single GPU; Oxford developed a virtual assistant that cut the human review time of supernova signals by 85% with a model on one physicist’s laptop; and Birmingham’s AI “emulators” help evaluate complex star evolution models. The timing of these tools couldn’t be better: Chile’s Vera Rubin Observatory just began capturing “the largest astronomical movie of all time” by taking an image of the universe roughly every 40 seconds for the next 10 years.
• IBM unveiled the world’s first sub-1-nanometer chip technology. The chip is slightly larger than a fingernail and packs nearly 100 billion transistors—almost double the density of its predecessor. The new chip runs up to 70% more efficiently, according to IBM’s technical report, thanks to a new “nanostack” architecture that “vertically stacks and staggers transistors.” The tech isn’t yet ready for manufacturing and is still roughly five years from market, the company said. But when it does enter production, your personal devices could eventually run more powerful apps and include new features without draining the battery or getting bigger. —LC