On July 1, 2026, Anthropic redeployed Claude Fable 5, its top-tier model that had been paused under export controls. After applying a strengthened cybersecurity classifier, users are reporting refusals and degraded performance on everyday tasks such as coding.
July 1, 2026 · Anthropic
Claude Fable 5 returns — but developers say the "relentlessly proactive" model came back nerfed
After an export-control suspension, Anthropic redeployed its flagship with "expanded safety margins." It insists the model is unchanged — but stricter safety classifiers now trigger constantly, and benchmarks have cratered.
50%
capped at half the weekly limit until a paid credit system on July 7
1M
token context window on the relaunched model
18days
from June 12 halt to June 30 lifting of restrictions
BridgeBench: initial release vs. after relaunch
Community benchmark scores — taller column is better.
Anthropic: "the model itself is unchanged"
Only "updated guardrails with expanded safety margins" were applied. But developers report the stricter classifiers over-trigger and force a visible fallback to Opus 4.8.
Trigger keywords
False positives
Obvious shift
Still powerful when it works
Tasks completed without refusal retain performance comparable to the initial release. Future updates are expected to bring improvements, with AWS, Google Cloud and Microsoft Foundry availability planned.
The trade-off
Export controls hit days after launch amid jailbreak concerns. Tightened safety guardrails now surface a clear tension between safety margins and real-world coding performance.
Reference API pricing
$10 / M input tokens
$50 / M output tokens
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…