The Mythos 5 Ban Is Creating an Asian AI Security Boom
- What happened
- 360 Security launched two AI cyber tools (Tulongfeng and Yitianzhen) as Japan's Sakana AI released Fugu — a multi-agent model marketed as 'ban-proof' — filling the gap left by the Mythos 5 export ban.
- Why it matters
- The two-week ban is fragmenting the AI security market along geographic lines, creating regional winners while US access remains restricted to ~100 vetted organizations even after partial restoration.
- What to do
- If you're evaluating AI security tools outside the US, add Sakana Fugu to your shortlist — but weigh routing transparency, 4–6× token overhead, and regional availability alongside benchmark scores.
Verdict: The Mythos 5 export ban is reshaping the AI security market faster than expected — and the winners aren't American. Two weeks after the June 12 export directive blocked global access, Chinese firm 360 Security and Japan's Sakana AI have launched credible alternatives that don't need US government approval. The market is fragmenting along geographic lines, and that fragmentation looks permanent (NBC News, 2026; TechCrunch, 2026).
What happened
Chinese cybersecurity firm 360 Security unveiled two AI-powered tools this week. Tulongfeng is a vulnerability-discovery model — founder Zhou Hongyi called it a "national strategic asset," signaling that China views this as defense-industry infrastructure, not just a commercial play. Yitianzhen handles automated cyber defense and incident response (TechCrunch, 2026).
On the same Monday, Tokyo-based Sakana AI launched Fugu, a multi-agent orchestrator that matches frontier models on SWE-Bench Pro (73.7), LiveCodeBench (93.2), and GPQA-D (95.5). Sakana's website now advertises "delivering frontier capability without the risk of export controls" — a direct pitch to enterprises shut out of Mythos 5 access. Co-founder David Ha described orchestration models as "the next frontier, beyond bigger models," arguing that "access to top models can disappear overnight" (TechCrunch, 2026).
On Friday, the US government partially reversed course — granting Anthropic permission to restore Mythos 5 to approximately 100 vetted US organizations, including government agencies and Fortune 500 companies (NBC News, 2026). For the rest of the world, the ban holds.
Why it matters
Three dynamics are now reshaping the AI security market:
- Geographic specialization is accelerating. 360 Security builds for China's defense sector. Sakana targets the broader Asian enterprise market. Neither competes with Anthropic globally — they're carving regional strongholds the ban manufactured.
- "Ban-proof" is now a product feature. Sakana's positioning is novel: export-control immunity marketed as a competitive advantage. For enterprises in countries worried about becoming the next target of US export restrictions, "this model will never be banned" is a pitch Anthropic and OpenAI cannot make — and it's resonating.
- The AI security category expanded from two players to five. Before June 12, AI-powered vulnerability discovery and cyber defense were dominated by Mythos 5 (83.8% CyberGym) and GPT-5.5-Cyber (85.6% CyberGym) — both restricted to vetted US defenders. Now Fugu, 360 Security's two tools, and the incumbents are competing in a suddenly crowded field.
Sakana's "ban-proof" positioning is a genuine differentiator — but Fugu's opaque routing (you can't see which underlying model executed your query) and 4–6× token overhead are real tradeoffs. The 360 Security tools are too new to assess — we haven't tested them, and the company hasn't responded to press inquiries about international availability.
What changes for you
The AI security tooling landscape just got more complex. You now have to weigh five dimensions:
| Dimension | Mythos 5 | Sakana Fugu | 360 Security | GPT-5.5-Cyber |
|---|---|---|---|---|
| Access | ~100 US orgs only | Open API (excl. EU/EEA) | China-focused | Vetted US defenders |
| Regulatory risk | High — subject to export controls | Low — ban-proof by design | Uncertain | Moderate — US-based |
| Transparency | Full model access | Opaque routing | Unknown | Full model access |
| Token cost | $10/$50 per 1M | $5/$30 (standard) | Not disclosed | Trusted Access (N/A) |
| Benchmark (CyberGym) | 83.8% | Not applicable (orchestrator) | No public scores | 85.6% |
If you're outside the US evaluating AI security tools: add Fugu to your shortlist, but inspect routing overhead and transparency before committing. If you're among the ~100 with restored Mythos 5 access: competitive pressure from Asian alternatives should accelerate features and pricing. Either way, US export-control risk is now a permanent procurement criterion — plan accordingly.
FAQ
Is Sakana Fugu a direct replacement for Mythos 5? No. Fugu is a multi-agent orchestrator that routes queries across a pool of expert models, while Mythos 5 is a single frontier model purpose-built for cybersecurity. Fugu matches Mythos-level benchmarks by orchestrating weaker models — it's a different architecture with different tradeoffs (opaque routing, 4–6× token overhead).
Will the export ban last? Partial restoration to ~100 US organizations is now underway (NBC News, 2026), but global access remains blocked with no timeline. Sakana co-founder Ren Ito argued at the G7 summit that the US should "preserve access" for allies, but the administration has not signaled a broader reversal. The longer non-US access stays blocked, the more entrenched Asian alternatives become.
Can I use 360 Security's tools outside China? Currently unclear. The tools were unveiled with defense-industry framing ("national strategic asset"), and 360 Security did not respond to TechCrunch's request for comment. For now, these are effectively China-market products.
What to do
- 1 If you're outside the US, add Sakana Fugu to your AI security evaluation shortlist
- 2 Audit your current AI security stack for US export-control dependency
- 3 Weigh routing transparency and token overhead alongside benchmark scores when evaluating orchestrator-based security tools
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.