Podcast Episode
The trigger appears to have been a jailbreak. A rival reportedly claimed it had bypassed Mythos's safeguards to unlock its cybersecurity capabilities. Anthropic countered that the jailbreak was narrow and worked in only one specific instance, but the government acted on national security grounds anyway. Crucially, the restriction hit only the top tier: everyday Claude models, including Claude Opus 4.8, kept running normally.
In Tokyo, Sakana AI took a different route. Its Fugu and Fugu Ultra models, released to general availability on 22 June, act as a "conductor" that dynamically routes tasks across a pool of frontier models, including GPT-5.5, Gemini 3.5 Flash and Claude Opus 4.8, through a single API. Sakana claims Fugu is "shoulder-to-shoulder with Fable 5 and Mythos" on engineering, scientific and reasoning benchmarks, while "delivering frontier capability without the risk of export controls."
Asian AI Firms Launch Mythos Alternatives as US Export Ban on Anthropic Eases
June 30, 2026
0:00
6:17
Two Asian AI companies, China's 360 Security Technology and Japan's Sakana AI, have launched products pitched as alternatives to Anthropic's banned Mythos 5 and Fable 5 models. The launches followed an 11 June US Commerce Department export directive, which was partially lifted on 27 June to restore Mythos 5 access for over 100 companies.
A frontier model locked down at home
In one of the stranger episodes in AI policy this year, Anthropic spent June unable to let most of the world use its two most advanced models. On 11 June, the US Commerce Department issued an export control directive forcing the company to disable all access to Mythos 5 and Fable 5 for any foreign national, including those inside the United States and even Anthropic's own non-citizen employees. The company argued the order's breadth left it no choice but to switch the models off for everyone.The trigger appears to have been a jailbreak. A rival reportedly claimed it had bypassed Mythos's safeguards to unlock its cybersecurity capabilities. Anthropic countered that the jailbreak was narrow and worked in only one specific instance, but the government acted on national security grounds anyway. Crucially, the restriction hit only the top tier: everyday Claude models, including Claude Opus 4.8, kept running normally.
Asia moves to fill the gap
The suspension left a vacuum, and two Asian firms moved fast. At the ISC.AI 2026 conference in Beijing, Chinese cybersecurity company 360 Security Technology unveiled Tulongfeng, which founder Zhou Hongyi called "China's version of Mythos." Built for automated vulnerability discovery, 360 says it has already flagged 3,432 software flaws, 105 of them confirmed by Chinese authorities. A companion system, Yitianzhen, handles automated cyber defence. Zhou conceded a 20-30% capability gap behind US models, but argued for a professional attack-and-defence team over reliance on a single advanced AI.In Tokyo, Sakana AI took a different route. Its Fugu and Fugu Ultra models, released to general availability on 22 June, act as a "conductor" that dynamically routes tasks across a pool of frontier models, including GPT-5.5, Gemini 3.5 Flash and Claude Opus 4.8, through a single API. Sakana claims Fugu is "shoulder-to-shoulder with Fable 5 and Mythos" on engineering, scientific and reasoning benchmarks, while "delivering frontier capability without the risk of export controls."
The sovereign AI scramble
Both launches reflect a wider regional push toward sovereign AI and reduced dependence on American providers. The story is still moving: on 27 June the US partially reversed course, restoring Mythos 5 access to more than 100 companies, though Fable 5 remains restricted. The episode has shown rivals and allies alike why they might want frontier capability that no single government can switch off.Published June 30, 2026 at 12:55am
