Observed Signal · Jun 29, 2026 · Product Launch · Source: t3n · Impact: 3/5 · Sentiment: Neutral
Asian AI Models Tulongfeng and Fugu Challenge Mythos 5
The Wall Street Journal reports that Chinese AI models are closing the performance gap with leading US systems in cybersecurity tasks. Security researchers told the WSJ that Zhipu AI’s GLM-5.2 can match US models at finding software vulnerabilities and in some Semgrep tests even outperformed Anthropic’s Claude Opus 4.8; with additional instructions both Opus 4.8 and GLM-5.2 reached performance comparable to Anthropic’s Mythos. GLM-5.2 is an open-weight model that can be downloaded and run on private hardware. Chinese firms are also releasing specialized tools — 360 Security Technology unveiled the Tulongfeng vulnerability-finding model — and Microsoft is reportedly considering offering Chinese models on its platforms. The article notes US restrictions on access to top Western models (OpenAI’s GPT-5.6 access limits) amid concerns about national security and competitive dynamics.
The story shows how US export controls on advanced AI (Anthropic's Mythos 5) are driving national alternatives in China and Japan; that shift affects global access to foundational models and the competitive AI supply landscape relevant to technology-dependent industries.
Track Anthropic Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- The Wall Street Journal reported insiders saying Chinese AI models now rival Anthropic’s Mythos in cybersecurity tasks.
- Zhipu AI’s GLM-5.2 was tested by cybersecurity firm Semgrep and sometimes outperformed Anthropic’s Claude Opus 4.8 in vulnerability detection.
- GLM-5.2 is an open-weight model that can be downloaded, run on private hardware, and freely adapted.
- 360 Security Technology announced the vulnerability-detection model Tulongfeng and said it reaches Mythos-level performance.
- Microsoft is reportedly evaluating whether to offer Chinese AI models on its platforms; OpenAI has restricted access to its GPT-5.6 model at the US government's instruction.
Connected Companies & Entities
7 Entities mapped“Anthropic released Mythos 5 and the stripped-down version Fable 5 and subsequently had to shut them down after the US government imposed an ...”
“The AI startup Sakana AI says its model Fugu can stand up to Fable 5 and Mythos Preview and that the company had been working on the model s...”
“The article references reporting by Reuters for details such as 360's claims and Zhou Hongyi's comments....”
“The article references reporting by Reuters for details such as 360's claims and Zhou Hongyi's comments....”
“The article references TechCrunch reporting about Sakana AI's timing and positioning of Fugu amid the US export ban....”
“The content is published on the German technology publisher t3n and reports on the Asian AI models and the Anthropic export-ban context....”
“The page notes it displays external editorial content from TargetVideo GmbH as a supplement to t3n's editorial offering....”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Asian Startups Launch Mythos-Like AI Models
Two Asian AI companies — Tokyo-based Sakana AI and Chinese cybersecurity firm 360 — this week launched new large AI models (Sakana’s Fugu and 360’s Tulongfeng and Yitianzhen) positioned as local alternatives to Anthropic’s Mythos/Fable after a U.S. export restriction limited Anthropic’s global access. Sakana presents Fugu as a frontier, agent-capable model tailored for Japanese language and customers, while 360 released vulnerability-discovery and cyber‑defense tools. The moves come amid concerns that U.S. export controls can remove access to leading models overnight, prompting regional providers to emphasize orchestration and local-language optimisation as hedges against concentrated AI supply. Anthropic reported a run‑rate revenue above $47 billion in May 2026, underscoring the commercial scale of the models affected by the export order.
White House AI Crackdown Helps Chinese Models Close Gap
The White House's export-control actions temporarily limited access to leading U.S. AI models, notably constraining Anthropic and prompting OpenAI to slow rollout of new models after a government request. Regulators allowed Anthropic to release Mythos 5 to select companies and agencies following a two-week shutdown, while Fable 5 remains unavailable. Meanwhile, Chinese firms such as Zhipu have released open-weight models (GLM 5.2) that researchers say match U.S. frontier labs on some cybersecurity benchmarks. Cost and efficiency advantages have prompted some startups and large firms to test or adopt Chinese open-weight alternatives, raising commercial interest but also cybersecurity concerns about models that can be run on-premise. The policy move could accelerate competitive shifts in AI development, enterprise sourcing and national-security debates over model and chip exports.
Anthropic disables Fable 5 and Mythos 5
On 12 June 2026 a U.S. government export-control directive instructed Anthropic to cut access to Claude Fable 5 and Mythos 5 for foreign nationals; Anthropic complied and suspended access globally, disrupting users and downstream products. The author reports a multi‑day, multi‑agent refactor run was interrupted when the model disappeared overnight, illustrating the operational risk of depending on hosted frontier models. Anthropic says it disagrees with the directive and is attempting to restore access. The incident sparked broader debate about “model sovereignty” and geopolitical risk for closed, hosted models, arriving alongside momentum in open-weight model releases and infra benchmarking. The piece uses prior platform shutdowns (Twitter, Reddit, Facebook, Shopify) as precedents and argues that open‑source weights offer stronger control, while recommending teams keep orchestration, prompts, data, and eval under their own control so a vendor pullback is an annoyance, not a business-stopping event.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.