Anthropic Redeploys Claude Fable 5 on July 1 After US Export Controls Lift, Adds New Cybersecurity Classifier

Anthropic is redeploying Claude Fable 5, its most succesful usually obtainable mannequin. On June 30, it announced that US export controls had lifted. The controls had lined Claude Fable 5 and Claude Mythos 5. Fable 5 returned to customers globally on Wednesday, July 1. Mythos 5 entry is restored to a set of US organizations.

The fashions have been pulled on June 12. A US authorities directive restricted them to non-foreign-nationals. Anthropic couldn’t confirm nationality in actual time. So it suspended each fashions for everybody.

This article explains what triggered the block. It covers the brand new safeguard and the proposed jailbreak framework. It additionally reveals how Fable 5 compares to rivals like GLM-5.2.

Quick info

• Model: Claude Fable 5 (a Mythos-class mannequin made protected for common use)
• Event: Redeployed July 1, 2026 after export controls lifted
• Reason for pause: An Amazon report on a safeguard bypass
• Fix: A brand new security classifier that blocks the reported approach
• Pricing: $10 per million enter tokens, $50 per million output tokens
• Where: Claude Platform, Claude.ai, Claude Code, Claude Cowork

What occurred: the timeline

Anthropic launched Fable 5 and Mythos 5 on June 9. Both share the identical underlying mannequin. Fable 5 ships with robust safeguards for common use. Mythos 5 has some safeguards lifted for defensive cybersecurity companions.

On June 12, the US authorities utilized export controls. The order took impact instantly. Anthropic suspended entry reasonably than threat non-compliance.

The set off was a report from Amazon researchers. They discovered a technique of bypassing Fable 5’s safeguards. The immediate made the mannequin determine plenty of software program vulnerabilities. In one case, it produced code displaying methods to exploit one vulnerability.

By June 26, the federal government authorized restoring Mythos 5 for some US organizations. On June 30, the controls have been absolutely lifted.

Why Anthropic says the discovering was not distinctive

Anthropic examined whether or not the discovering was distinctive to Fable 5. It was not.

Less succesful fashions recognized the identical vulnerabilities. That record contains Claude Opus 4.8, GPT-5.5, and Kimi K2.7.

For the one exploit demonstration, each examined mannequin reproduced it. That set included Haiku 4.5, Sonnet 4.6, Opus 4.6, and Opus 4.7. It additionally lined Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7.

The Anthropic staff states the approach uncovered no distinctive Mythos-level cyber capabilities. It known as the case a borderline one for Fable 5’s safeguards. The blocked conduct concerned solely routine defensive cybersecurity work.

How the brand new classifier works

Anthropic nonetheless moved to shut the hole. It educated an improved security classifier for the reported conduct.

The classifier blocks the precise approach in over 99% of circumstances. Blocked requests will not be refused outright. They are routed to Claude Opus 4.8 as an alternative. Users are notified when this fallback occurs.

Researchers from the Department of Commerce’s CAISI examined each previous and new safeguards. They agree the safeguards are terribly robust. The tradeoff is extra false positives throughout routine coding and debugging.

This displays Anthropic’s ‘protection in depth’ design. Classifiers are smaller AI techniques that detect dangerous cyber duties. A deliberate ‘security margin’ additionally blocks some benign requests. Fable 5 makes use of a a lot bigger security margin than prior fashions.

The proposed jailbreak severity framework

The episode uncovered a niche. The business has no shared normal for scoring a ‘jailbreak,’ a method that bypasses a mannequin’s safeguards.

Anthropic is drafting one with Amazon, Microsoft, Google, and different Glasswing companions. The draft scores a jailbreak on 4 standards:

1. Capability achieve — how far past current instruments it takes the person.
2. Breadth of functionality achieve — what number of distinct offensive duties it unlocks.
3. Ease of weaponization — how a lot human effort an assault nonetheless wants.
4. Discoverability — how simply somebody can acquire the approach.

For essentially the most extreme class, Anthropic will deploy preliminary mitigations instantly. It can also be standing up 24/7 monitoring of jailbreak submission channels.

Interactive scorer

Try this embedded interactive scorer to see how these 4 standards mix.

(*1*)

Use circumstances with examples

Fable 5 targets long-horizon, agentic work. Here is the place early engineers can apply it.

• Codebase migrations: Stripe reported a codebase-wide migration in at some point. The job spanned a 50-million-line Ruby codebase. Doing it by hand would take a staff over two months.
• Financial evaluation: On Hebbia‘s Finance Benchmark, Fable 5 posts the best rating. It features on chart, desk, and doc reasoning.
• Vision-to-code: Fable 5 can rebuild an internet app’s supply code from screenshots alone.
• Long-running brokers: File-based reminiscence helps it keep centered throughout tens of millions of tokens.

How Fable 5 compares

The pause created a gap for rivals. Days after the suspension, Zhipu AI released GLM-5.2 as open weights. Independent testers rank it the strongest brazenly obtainable mannequin.

GLM-5.2 makes use of a Mixture-of-Experts design. It has roughly 750 billion whole parameters. Only about 40 billion activate per token. On Semgrep‘s IDOR benchmark, it scored 39% F1. That beat Claude Code at 32% on the identical immediate.

The hole narrows on price. On AA-Briefcase, Fable 5 averaged $31 per job. GLM-5.2 averaged $2.40.

Access and a fast API instance

For Pro, Max, Team, and choose Enterprise plans, Fable 5 is included via July 7. It covers as much as 50% of weekly utilization limits. After that, entry strikes to usage credits. Anthropic can also be re-enabling Fable 5 on AWS, Google Cloud, and Microsoft Foundry.

Developers name the mannequin with the claude-fable-5 string:

from anthropic import Anthropic

# Reads your key from the ANTHROPIC_API_KEY setting variable
consumer = Anthropic()

message = consumer.messages.create(
    mannequin="claude-fable-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Refactor this module for readability."}
    ],
)

print(message.content material)

If a classifier fires, the response comes from Opus 4.8. Your code path stays the identical.

Key takeaways

• Fable 5 returns July 1 after export controls have been lifted.
• A brand new classifier blocks the reported bypass in over 99% of circumstances.
• Blocked requests path to Opus 4.8, not an outright refusal.
• Anthropic proposes a four-criteria framework for scoring jailbreaks.
• GLM-5.2 emerged as a less expensive open-weight rival throughout the pause.

Check out the Technical details. Also, be happy to observe us on Twitter and don’t neglect to hitch our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to accomplice with us for selling your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar and so forth.? Connect with us

The submit Anthropic Redeploys Claude Fable 5 on July 1 After US Export Controls Lift, Adds New Cybersecurity Classifier appeared first on MarkTechPost.