Breaking

Anthropic and White House stay deadlocked over Claude Fable 5 export curbs

June 16, 2026 at 03:16 EDT

Foundation Models
Policy & Regulation
Security

Anthropic publicly released its most advanced model, "Claude Fable 5," with added safeguards in June 2026, only to be ordered by the US government to cut off all foreign-national access shortly after, and talks in Washington DC have so far failed to lift the export controls.

June 16, 2026 · Anthropic vs. White House

Talks Collapse Over "Claude Fable 5" — Export Controls Stay in Place

Washington negotiations ended without resolution. The administration keeps Commerce Department export controls active, citing serious cyber risk if the model's safety guardrails can be jailbroken.

June 9

Fable 5 & Mythos 5 released (Mythos-class)

June 12

Commerce export controls; both models disabled

June 16

Talks conclude — controls NOT lifted

API PRICING vs. OPUS 4.8 — roughly 2× the cost

Per 1M tokens · taller column = higher price

$5 / $25 (est. Opus 4.8)

out

Opus 4.8

$10 / $50

out

Fable 5

token context window (128K max output)

<5%

of sessions trigger guardrails → falls back to Opus 4.8

SOTA

top scores in engineering, vision & science

HOW THE GUARDRAIL WORKS — and where it breaks

Risky prompt
(cyber / bio / chem)

→

Safety classifier fires

→

Switch to Opus 4.8
lower-tier model

→

JAILBREAK bypasses it
crux of dispute

Anthropic's position

The model's risks are "narrow and non-universal." Wants controls lifted so the highest-performance model can ship widely.

Administration's position

Cyber capabilities exposed by bypassing guardrails are "serious." Export controls stay; access stays limited.

Continue reading

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