Chinese cybersecurity firm 360 has unveiled an artificial intelligence tool named "Tulongfeng," which the company claims can go head-to-head with Anthropic’s Mythos. This announcement arrives amid the US export ban.
How does Tulongfeng compare to Western models?
360 introduced Tulongfeng as a specialized cybersecurity-focused AI. The company asserts that the model performs at a level competitive with Anthropic’s Mythos and Fable 5 models.
360’s marketing strategy focuses on positioning its tool as a sovereign alternative that operates independently of the restricted hardware and software ecosystems governed by U.S. trade policies. The tool is designed to assist in threat detection, vulnerability analysis, and automated security response, areas where Chinese firms are currently prioritizing AI integration to maintain network defense capabilities.
Why are Chinese AI firms shifting strategy?
The push for independent AI development follows the U.S. These sanctions have made it increasingly difficult for Chinese labs to train large-scale models on the same hardware footprints used by U.S.-based competitors.
In response, companies like 360 and the Tokyo-based startup Sakana AI—which has also been noted for its research into automated model evolution—are focusing on architectural efficiency. By optimizing models to run on domestic compute clusters or through specialized, smaller-parameter designs, these firms aim to bypass the performance bottlenecks created by the lack of cutting-edge hardware.
What are the risks of specialized AI models?
Industry analysts note that while domestic models like Tulongfeng may demonstrate high proficiency in specific tasks like cybersecurity, they face significant hurdles regarding general-purpose reasoning and data diversity.
- Hardware Constraints: The reliance on older-generation or domestic-made chips increases the cost and time required to fine-tune these models.
- Regulatory Compliance: Both Chinese and U.S. developers must navigate complex regulatory landscapes. In China, the Cyberspace Administration of China (CAC) requires that generative AI models undergo security assessments and adhere to strict content moderation guidelines before they are deployed for public or commercial use.
- Performance Benchmarking: Independent verification of 360’s claims remains limited. Because the company’s performance benchmarks are internal, third-party researchers have yet to confirm how Tulongfeng stacks up against global benchmarks.
Looking ahead
The development of Tulongfeng signals a broader trend in China’s tech sector: moving away from chasing massive, general-purpose models toward creating highly optimized, vertical-specific AI. As the U.S. continues to restrict access to advanced computing power, the effectiveness of these specialized tools will likely determine the competitiveness of Chinese firms in the global cybersecurity and software markets. 360 has not yet announced a global release date for the tool, keeping its focus on domestic enterprise and government clients.