Anthropic Seeks AI-Blocking Powers, Gets Blocked Instead

June 13, 2026·AI Newsroom·4 min read·6 sources
anthropicpolicyregulationfable-5mythos-5export-control+1high-risk claims
Anthropic policy framework illustration
Image: Anthropic / Advanced AI Framework policy page (June 10, 2026)

On 2026-06-10, Anthropic published a two-part policy proposal — the Advanced AI Framework and an Economic Policy Framework — formally asking the US federal government for “the legal authority to block or deter the deployment of models that pose a significant risk of catastrophic harm”, with graduated civil penalties tied to global revenue (Anthropic, June 10, 2026). Two days later, an export control directive ordered the immediate suspension of access to Claude Fable 5 and Claude Mythos 5 — Anthropic’s flagships, launched three days earlier — for all customers, not just foreign ones. Anthropic complied, then cited its own proposal to challenge the process (Anthropic, June 12, 2026).

It’s a 48-hour arc, signed by one actor on both sides: a major lab requesting regulatory power over its own models, and that power being exercised against its models without following the process the lab had invoked.

What happened

June 10 — the proposal. The Advanced AI Framework defines a precise frontier threshold: models trained with more than 10²⁵ FLOP, developed by companies with over $500M in AI revenue or $1B in AI R&D. It names four catastrophic risk categories — biological, cyber, loss of control, and automated R&D that accelerates the other three — and asks for civil penalties scaled to global revenue for repeated violations (Anthropic, June 10, 2026). CEO Dario Amodei’s same-day essay frames the model in the US regulatory tradition — “airplanes, not nuclear weapons” — and cites the June 2026 Trump Executive Order as incremental progress, recommending more action (darioamodei.com, June 2026).

The proposal isn’t a blank check. The PDF specifies that blocking authority should be “scoped to the above four specific risks” with “protective measures against political favoritism or arbitrary decisions” and that the process must be “statutory, transparent, fair, clear, and grounded in technical facts” (Advanced AI Framework, June 10, 2026). These are the words the June 12 statement will reuse — and contest.

June 12 — the directive. At 5:21 PM ET on June 11, Anthropic received an export control directive ordering immediate suspension of Fable 5 and Mythos 5 access for any “foreign national,” including Anthropic employees of foreign nationality. To ensure compliance, Anthropic disabled both models for all customers. Other Claude classes (Sonnet, Opus, Haiku) were not affected (Anthropic, June 12, 2026).

The government letter, per Anthropic, “did not provide specific details of its national security concern.” Anthropic’s understanding is that the government learned of a method to bypass Fable 5’s safeguards. Anthropic reviewed a demonstration and called the results “a small number of previously known, minor vulnerabilities,” replicable “by other publicly-available models … without requiring a bypass” — pointing explicitly at OpenAI GPT-5.5, whose system card classifies it as “High capability in the Cybersecurity domain, but below Critical” (OpenAI GPT-5.5, 2026-04-23).

Anthropic’s closing line is direct: “We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.” (Anthropic, June 12, 2026).

Why it matters

1. The first operational precedent of blocking power without the statutory process requested. Fable 5 launched on June 9, 2026 as the first publicly available “Mythos” class model. Three days later it’s unreachable by any customer — the first time the suspension power is exercised against the lab that had publicly requested that power. Anthropic’s own framing: “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers” (Anthropic, June 12, 2026).

2. The “all customers” cutoff is operational, not semantic. The directive prohibits access to foreign nationals; for compliance, Anthropic disabled access for everyone. Any team running eval, logs, or agents on claude-fable-5 or claude-mythos-5 needs a fallback. Sonnet, Opus, and Haiku users are unaffected on the API — but the regulatory precedent is.

3. The policy artifact of the week isn’t the statement — it’s the proposal. The Advanced AI Framework is the most detailed AI policy proposal ever published by a major lab: 27 pages with measurable compute thresholds, measurable applicability criteria, graduated penalties, and explicit safeguards against misuse. The real question isn’t what happens to Fable 5; it’s what happens to the relationship between blocking power and statutory process in the next twelve months.

What to watch

  1. The full text of the export control directive (or an unclassified version).
  2. OpenAI’s response to the GPT-5.5 equivalence claim — Anthropic’s case depends on it.
  3. Statements from Google DeepMind, Meta, xAI, Microsoft AI.
  4. The relationship between the June 2026 Executive Order and the June 12 directive.
  5. Fable 5 and Mythos 5 restoration date (Anthropic: “as soon as possible”, no date).
  6. The Advanced AI Framework’s trajectory on Capitol Hill and in Brussels over 3–6 months.

Risks and caveats

Verdict

A 48-hour arc: a major lab publicly requests stricter rules for itself, those rules become operational in 48 hours in a form the lab contests. The open question isn’t when Fable 5 and Mythos 5 return — it’s whether the regulatory precedent holds, and whether other labs join Anthropic’s procedural position or distance themselves from it.

“We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.”

— Anthropic, statement of June 12, 2026 (Anthropic)