AI

Asian AI startups launch Mythos-like models as Anthropic’s export ban drags on

New AI models from Sakana AI and China’s 360 show why access to frontier capability is becoming an operational dependency, not just a technology preference.

Why this story matters

TechCrunch reports that AI companies in China and Japan are using the current US restrictions on Anthropic’s Mythos and Fable models as an opening to promote alternative systems. Chinese cybersecurity firm 360 has reportedly unveiled Tulongfeng, a tool it says can compete with Mythos in vulnerability discovery. Tokyo-based Sakana AI has launched Fugu, a model it says can stand alongside Anthropic’s Fable 5 and Mythos Preview.

The story is not only about geopolitics. It is about operational dependency. If access to a frontier model can disappear or become restricted, businesses and public bodies that have built critical processes around a single supplier need to know what happens next.

What stands out

Sakana is positioning Fugu as a hedge rather than a clean break from US models. The company told TechCrunch that US models remain important to Asia, but it also describes Fugu as a way to reduce exposure to tightening export controls. Its message is blunt: access matters, and relying on one provider for important infrastructure is risky.

360’s positioning is more assertive. According to TechCrunch, the Chinese firm has presented vulnerability-finding AI as a national strategic asset. It has also reportedly launched Yitianzhen, a tool aimed at automating cyber defence and incident response. That combination matters because cybersecurity AI is not just another productivity feature. It can change both attack and defence capabilities.

What leaders should take from this

Most organisations will not be choosing between Mythos, Fugu and Tulongfeng this week. That is not the useful lesson. The useful lesson is that AI availability, jurisdiction, language fit, export controls and supplier concentration are becoming part of the operating model.

A business using AI in important workflows should be able to answer some basic questions:

  • Which processes depend on one model provider?
  • What happens if that provider becomes unavailable, restricted or commercially unattractive?
  • Which use cases are security-sensitive enough to need extra review?
  • Who owns the fallback plan?
  • What evidence would show that an alternative model is safe enough for the job?

Why this matters in practice

The early AI adoption conversation was full of demos and model comparisons. This story points to a more serious phase: access, resilience and control. Capability is spreading, but not evenly, and not always through the suppliers organisations expected to rely on.

For DTP’s clients, the practical response is not to chase every new model. It is to map the workflows where AI dependency is becoming real, decide which dependencies are acceptable, and put fallback routes in place before a supplier or policy shock forces the issue.