US Forces Anthropic to Disable Its Flagship Models + OpenAI Faces 42-State Probe — June 13, 2026

⚡ Top Story

US Commerce Department Orders Anthropic to Disable Fable 5 and Mythos 5 for All Foreign Nationals — Citing a Jailbreak It Won't Detail

Anthropic disclosed on June 13 that it received a directive from the US government on June 12 at 5:21pm ET — a letter from Commerce Secretary Howard Lutnick, drafted with the Commerce Department's Bureau of Industry and Security — ordering it to immediately suspend all access to Claude Fable 5 and Mythos 5 for any foreign national, whether inside or outside the US, including Anthropic's own foreign-national employees. The directive invokes export-control authority and cites an unspecified national-security concern. Anthropic says its understanding is that the government became aware of a jailbreak technique against Fable 5, but Anthropic believes the exploit was narrow — unlocking one specific capability rather than defeating the model's safeguards universally. The company says it is complying with the legal order while publicly disagreeing with the rationale. Access to all other Anthropic models is unaffected.

Why it matters: This is the first known case of the US government using export-control authority to force a frontier AI lab to abruptly pull a flagship model from foreign users — just three days after Fable 5 and Mythos 5's public launch (covered June 10) and as Anthropic prepares for a blockbuster IPO. It sets a precedent: government national-security review can now reach into a model's access controls after release, not just its training-time export classification, and on a timeline measured in hours.

Sources: Anthropic Newsroom · Bloomberg · NBC News · Fortune

🔬 Research & Papers

Nothing independently verifiable and genuinely new surfaced from arXiv, major conferences, or research labs in this 24-hour window. (Google DeepMind's "From AGI to ASI" roadmap was submitted June 10 and is outside today's window.)

🏢 Industry & Startups

Nothing genuinely new beyond the stories covered in Government & Regulation and Tools & Releases surfaced in this window.

🛠️ Tools & Releases

Moonshot AI Open-Sources Kimi K2.7-Code — a 1-Trillion-Parameter Coding Model Undercutting GPT-5.5 and Claude on Price

On June 12, Moonshot AI published Kimi K2.7-Code to Hugging Face under a Modified MIT license — a Mixture-of-Experts model with 1 trillion total parameters (32B active) and a 256K context window, built on Kimi K2.6 and tuned specifically for long-horizon, agentic software engineering (multi-step planning, tool use, debugging across large repos) rather than general chat. Moonshot reports a +21.8% gain on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite versus K2.6, with 30% fewer reasoning tokens. It's priced at $0.95 / $4.00 per million input/output tokens — up to 12x cheaper than GPT-5.5 and Claude on a per-token basis, though it still trails them on standard coding benchmarks.

Why it matters: another data point in the steady cadence of Chinese open-weight labs (Moonshot, MiniMax, Alibaba) shipping frontier-adjacent, agent-oriented coding models at a fraction of Western API pricing — intensifying price pressure on closed-model providers for the high-volume "coding agent" workload specifically.

Sources: MarkTechPost · The Decoder · CryptoBriefing

🌏 Global AI & Geopolitics

Nothing additional beyond the Anthropic export-control story (Top Story) and the OpenAI state-AG probe (Government & Regulation) surfaced in this window.

⚡ Energy, Infrastructure & Chips

Nothing notable and independently verifiable surfaced specifically in this 24-hour window beyond ongoing threads already covered in recent briefings.

🤖 AI Agents & Autonomy

Nothing genuinely new and independently verifiable emerged in this window beyond Kimi K2.7-Code's agentic coding focus (covered in Tools & Releases).

🔒 Safety, Alignment & Ethics

The core safety angle of the day is the Anthropic story (Top Story): a government-cited jailbreak of Fable 5 led directly to a forced access suspension — a real-world case of a reported exploit triggering regulatory action within days, rather than a quiet patch.

📊 Numbers & Signals

5:21pm ET, June 12 — when Anthropic received the US export-control directive
42 — state attorneys general involved in the new investigation into OpenAI
1T / 32B — Kimi K2.7-Code's total vs. active parameters (MoE)
256K — Kimi K2.7-Code's context window
+21.8% — Kimi K2.7-Code's gain on Kimi Code Bench v2 over K2.6
$0.95 / $4.00 — Kimi K2.7-Code's price per million input/output tokens (up to 12x cheaper than GPT-5.5/Claude)

🧠 Worth Thinking About

Two of today's stories are really one story told from different angles: the US government simultaneously tightened its grip on a frontier lab from the outside in (OpenAI, via a 42-state consumer-protection subpoena probing how its products affect users) and from the inside out (Anthropic, via an export-control order reaching directly into a shipped model's access permissions). Both labs are heading toward historic IPOs within months of each other, and both are now learning that going public doesn't just invite financial scrutiny — it invites the government to treat their models' behavior and reach as matters of state interest, on compressed timelines (a few hours' notice for Anthropic, a Friday-afternoon subpoena for OpenAI). The era where frontier labs negotiated primarily with regulators about future rules seems to be giving way to one where the government acts on already-shipped products in near real time.

🏛️ Government & Regulation

Coalition of 42 State Attorneys General Opens Sweeping Investigation Into OpenAI, Days After Its Confidential IPO Filing

OpenAI was served Friday, June 12, with a subpoena from a coalition of 42 state attorneys general (led by New York's AG) seeking documents on a sweeping range of topics: advertising practices, user engagement and retention design, handling of consumer and health data, marketing and product design as it relates to minors and seniors, internal safety-testing policies before releases, and its deep-learning models generally. The probe lands days after OpenAI confidentially filed a draft S-1 for a US IPO (reported June 1, targeting as early as September at up to ~$1T valuation) and amid an existing Florida lawsuit alleging OpenAI misrepresented ChatGPT's safety. OpenAI said: "We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices."

Active compliance countdowns:

June 27, 2026 (14 days): GPT-4.5 retirement from ChatGPT
June 30, 2026 (17 days): Colorado's comprehensive AI law takes effect
August 2, 2026 (50 days): EU AI Act Article 50 transparency rules take effect
August 26, 2026 (74 days): o3 retirement from ChatGPT

Sources: Bloomberg · Engadget · The Next Web

🔭 Frontier Lab Dispatch

Anthropic — Statement on the US Government Directive to Suspend Access to Fable 5 and Mythos 5 (June 13, 2026)

Covered in full in the Top Story above — Anthropic's own account of the export-control order, its compliance, and its disagreement with the government's stated rationale.

Source: anthropic.com

No other verified, directly-sourced posts from OpenAI, Google DeepMind, Meta AI, or leading Chinese labs landed within this 24-hour window beyond the Kimi K2.7-Code release (Tools & Releases).

🔗 Quick Links

Tier 1 — Frontier AI Labs / Official

Anthropic: Statement on the US government directive to suspend access to Fable 5 and Mythos 5

Tier 2 — Business & Finance Media

Tier 3 — Tech & AI News Media