MRAJ @MRAJf2ts - Twitter Profile

MRAJf2ts retweeted

6 days ago

Claude Fable 5 is back online — worldwide, as of yesterday. On June 30, the U.S. Commerce Department lifted the export controls it had slapped on Fable 5 (and the tighter Mythos 5) on June 12. The trigger: Amazon researchers found a jailbreak prompt that got Fable 5 to flag software flaws and write proof-of-concept exploit code. Here's what actually happened: — The jailbreak worked on plenty of other models too, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7. Anthropic called it routine defensive security work, not a hidden super-capability. — The government saw it differently. Emergency export controls cut off both models to ANY foreign national — including Anthropic's own non-citizen staff. The company couldn't verify nationality in real time, so it shut both models down for everyone. — To get Fable 5 back, Anthropic trained a new classifier that catches the exact jailbreak technique. Stops it 99%+ of attempts. Blocked requests drop to the weaker Opus 4.8 instead. — Mythos 5 (fewer safety guardrails) stays on a shorter leash: ~100 critical-infrastructure companies and federal agencies only, for now. The bigger picture that nobody's talking about enough: Anthropic is now proposing a shared industry standard for ranking jailbreak severity — scored on capability gain, breadth, ease of weaponization, and discoverability. They've opened a HackerOne program. They're standing up a 24/7 jailbreak monitoring team. This is the first time export controls have been imposed on an AI model and then lifted. The precedent matters — and it sets a playbook for every future frontier-model incident. https://t.co/M8h7e5Sd9u

TeksCreate's tweet photo. Claude Fable 5 is back online — worldwide, as of yesterday.

On June 30, the U.S. Commerce Department lifted the export controls it had slapped on Fable 5 (and the tighter Mythos 5) on June 12. The trigger: Amazon researchers found a jailbreak prompt that got Fable 5 to flag software flaws and write proof-of-concept exploit code.

Here's what actually happened:

— The jailbreak worked on plenty of other models too, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7. Anthropic called it routine defensive security work, not a hidden super-capability.

— The government saw it differently. Emergency export controls cut off both models to ANY foreign national — including Anthropic's own non-citizen staff. The company couldn't verify nationality in real time, so it shut both models down for everyone.

— To get Fable 5 back, Anthropic trained a new classifier that catches the exact jailbreak technique. Stops it 99%+ of attempts. Blocked requests drop to the weaker Opus 4.8 instead.

— Mythos 5 (fewer safety guardrails) stays on a shorter leash: ~100 critical-infrastructure companies and federal agencies only, for now.

The bigger picture that nobody's talking about enough:

Anthropic is now proposing a shared industry standard for ranking jailbreak severity — scored on capability gain, breadth, ease of weaponization, and discoverability. They've opened a HackerOne program. They're standing up a 24/7 jailbreak monitoring team.

This is the first time export controls have been imposed on an AI model and then lifted. The precedent matters — and it sets a playbook for every future frontier-model incident.

https://t.co/M8h7e5Sd9u

0

1

0

253

MRAJ

@MRAJf2ts

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users