Skip to content
Submit incident
AII-1000-92677
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

U.S. imposes export controls on Anthropic AI models over security concerns

The U.S. Trump administration imposed export controls on Anthropic's advanced AI models Mithos 5 and Fable 5 after discovering that a South Korean telecommunications company with suspected ties to China was granted early access. Anthropic revoked the access after the disclosure, but the incident undermined government confidence in the company's security practices.

12 June 2026 ยท Anthropic ยท Governance, accountability and labour
12 June 2026
AII-1000-926772 sources โ†“
Quick facts
Category
Governance, accountability and labour
Country
United States
Source
๊ฒฝํ–ฅ์‹ ๋ฌธ
Date
12 June 2026
Drafted by
Raidu Incident Detection

What happened

Anthropic submitted a list of 111 organizations approved to access the Mithos 5 AI model before its official release through the Project Glasswing initiative. The Trump administration reviewed and approved the list weeks before June 16, 2026.

Anthropic later disclosed that approximately 50 additional organizations had been granted access to the model but did not initially reveal their identities. During a subsequent review, U.S. government officials identified a South Korean telecommunications company with suspected ties to China on the access list.

Anthropic revoked the South Korean telecommunications company's access shortly after the discovery. On June 12, 2026, the Trump administration announced export control guidelines blocking foreign individuals and institutions from accessing Mithos 5 and Fable 5, and Anthropic immediately suspended access for all users.

Reported impact

The discovery of access granted to an entity with suspected China ties undermined U.S. government trust in Anthropic's ability to protect sensitive technology, according to officials cited by the Washington Post.

Amazon reported security vulnerabilities in Anthropic's new models, which was cited as a contributing factor in the government's decision to impose export controls.

Index analysis

Project Glasswing was established by Anthropic to allow verified companies and institutions to preemptively access the Mithos model to identify and defend against software vulnerabilities before the system's official release.

Governance controls

The U.S. government required Anthropic to submit lists of approved users for access to Mithos 5 before release and reviewal the identities of access recipients.

Classification

Category
Governance, accountability and labour
Industry
Government, security and defence
Country
United States
Harm status
Potential harm
Intent
Unknown
Timing
Post-deployment
Harm types
Economic or property, Public interest
Harmed
Business, Government
Consequences
System suspension or termination
AI systems
Mithos 5, Fable 5
Regulations
EU AI Act (cited)

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

๊ฒฝํ–ฅ์‹ ๋ฌธ
์ค‘๊ตญ ์—ฐ๊ณ„ ์˜ํ˜น ํ•œ๊ตญ ํ†ต์‹ ์‚ฌ์— ๊ธฐ์ˆ  ์ œ๊ณต ์ด์œ ๋กœ ์•คํŠธ๋กœํ”ฝ ์ˆ˜์ถœํ†ต์ œ
16 June 2026
์—ฐํ•ฉ๋‰ด์Šค
"็พŽ์ •๋ถ€, ๋ฏธํ† ์Šค ์‚ฌ์ „์‚ฌ์šฉ ๋ช…๋‹จ์„œ 'ไธญ์—ฐ๊ด€์˜์‹ฌ' ้Ÿ“ํ†ต์‹ ์‚ฌ ๋ฐœ๊ฒฌ"
16 June 2026