Breaking

US Orders Anthropic to Pull Fable 5 Access After Amazon Flags Flaw

June 15, 2026 at 15:58 EDT

Security
Foundation Models
Policy & Regulation

On June 12, 2026, Anthropic cut off worldwide access for all users to its large language model "Claude Fable 5" and its underlying "Mythos 5" the same day it received a US government export-control directive, following warnings about a potential jailbreak that could elicit restricted cyberattack-related information. The matter originated with a discovery by Amazon researchers, after which the company's CEO Andy Jassy brought it to US officials and raised broad concerns about frontier-model capabilities.

Jun 12–13, 2026 · Frontier-Model Crackdown

Anthropic Halts Worldwide Access to Fable 5 & Mythos 5

After Amazon flagged a possible safety-guardrail bypass and CEO Andy Jassy raised it with US officials, the government imposed export-control measures — leading Anthropic to suspend both models for every user, including US citizens, just days after launch.

~3 days

From launch (~Jun 9) to suspension (~Jun 12–13)

2 models

Fable 5 & Mythos 5 both pulled worldwide

All users

Including US citizens, for compliance

How the suspension unfolded

1 · DISCOVERY

Amazon researchers find a prompt series that bypasses Fable 5's guardrails

→

2 · DISCLOSURE

Finding taken to Anthropic; Jassy then raises it with US officials

→

3 · RULING

Government bars foreign nationals from accessing both models

→

4 · HALT

Anthropic suspends worldwide access to comply

Two models, one halt

Fable 5

Safety-hardened, broad release
Strengthened guardrails (more refusals)
Shipped via Claude.ai / API / Bedrock
Access halted ~Jun 12–13

Mythos 5

Raw model, no safety classifiers
No guardrails
Restricted availability
Same halt

Defense

Anthropic disputes the "jailbreak" label, noting similar vulnerabilities exist in rival models like GPT-5.5. Supporters say it was singled out and that government intervention was excessive. Early reviews praised sustained performance on long, complex tasks.

Criticism

Officials weighed misuse risk heavily — including zero-day vulnerability discovery. Critics said Amazon "snitched" and that a safety-first stance backfired. Many users had complained Fable 5's guardrails triggered too often.

Restoration timing remains unclear — and the episode is likely to intensify the regulatory debate over frontier-model capabilities and national security.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.