The U.S. government has enacted unprecedented export controls on two advanced AI models from Anthropic. Millions of non-U.S. users lost access overnight, prompting Anthropic to publicly oppose the decision and warn of potential industry-wide repercussions.
U.S. Commerce Department Blocks Foreign Access to 'Fable 5' and 'Mythos 5'
The Trump administration issued export control directives to Anthropic, prohibiting foreign access to the Fable 5 and Mythos 5 models, regardless of location. The U.S. Commerce Department confirmed that it sent the directive, and Anthropic has complied by blocking access for all customers to the two models.
Anthropic received the directive at 5:21 PM Eastern Time on June 13, stating that the letter did not specify the security concerns involved. According to Anthropic, the government informed them that a 'jailbreak' method for Fable 5 had been identified, but Anthropic disagrees with the assertion that a verbal notification regarding a 'potentially narrow and non-universal jailbreak' technique justifies the service suspension.
Anthropic argued that the jailbreak represents a narrow vulnerability that activates cybersecurity capabilities in Mythos only under specific circumstances, and the same technique could apply to other publicly available models, such as OpenAI's GPT-5.5, which are not subject to similar export controls.
The company contended, "The discovery of a narrow potential jailbreak should not serve as grounds for recalling commercial models distributed to millions." They warned that if this standard is applied across the industry, it could effectively halt the rollout of new models by all frontier model operators. Fable 5 and Mythos 5, released on June 9, are Anthropic's latest high-performance models, noted for their exceptional cybersecurity vulnerability detection capabilities, which analysts believe contributed to this action.
HarmonyOS 7 Redesigned as 'Agentic AI' Operating System
Huawei officially unveiled HarmonyOS 7 at its developer conference. The operating system features the HarmonyOS Intelligent Agent Framework 2.0, promoting in-app command support and claiming over 90% task execution rates with its agentic AI assistant capabilities.
Huawei emphasized that HarmonyOS 7 lays the groundwork for transitioning to an agentic AI operating system. The key features include an agent-friendly system architecture, Agent Framework 2.0, and the full integration of the native AI assistant 'Celia.' A stable version of HarmonyOS 7 is expected to be released this fall. Amid tightening U.S. export regulations on AI, Huawei appears to be accelerating the development of its independent OS-based AI agent ecosystem.
Open Source Coding Agent Model 'Kimi K2.7-Code' Released
Chinese AI startup Moonshot AI has released Kimi K2.7-Code. This model, based on a mixture of experts (MoE) architecture, features a total of 1 trillion parameters with 32 billion active parameters and has been made open source under a Modified MIT license on Hugging Face. Compared to its predecessor K2.6, it reduces inference token usage by 30% while improving coding performance, designed for long-term autonomous coding workflows. As Chinese AI companies continue to unveil high-performance open-source coding models, competition with U.S. paid coding agents like Claude Code and OpenAI Codex is intensifying.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.