June 25, 2026, (Inside AI) — China must build its own version of Anthropic’s Mythos AI model to counter growing cybersecurity threats, according to Zhou Hongyi, founder of 360 Security Technology. Speaking at a Beijing cybersecurity conference on Wednesday, he described the US technology as a “cyber nuclear weapon” that cannot remain solely in American hands.
Mythos, released in April, autonomously identifies software flaws. Zhou said it accelerated vulnerability discovery a hundredfold while slashing costs. This “terrifying change” has “democratised” cyberattacks, he noted.
The US formed the Project Glasswing alliance in April. Over 40 organizations gained access to Mythos Preview to bolster defenses. China was excluded. Zhou warned this creates a strategic gap.
“This means US organisations can use Mythos to scan your vulnerabilities, but you don’t even have the qualification to catch a glimpse of Mythos,” Zhou said.
He called for a domestic equivalent. Such a system would serve as a “game-changing weapon in cyber warfare” and provide “reciprocal strategic deterrence capability” akin to nuclear-armed states.
Mythos and the Glasswing Alliance: A Strategic Lockout
Anthropic’s Mythos model marks a leap in automated vulnerability discovery. It scans codebases, spots weaknesses, and suggests fixes without human input. Early tests show it finds flaws in minutes that once took weeks. This efficiency reshapes cyber defense and offense.
Project Glasswing channels that power to a select group. Members include US agencies, critical infrastructure firms, and key allies. The alliance aims to harden networks before adversaries strike. But by locking out China, it deepens a digital divide.
Zhou’s nuclear analogy is deliberate. In Cold War logic, mutual vulnerability prevents first strikes. He argues a Chinese Mythos clone would restore balance. Without it, China’s networks remain exposed to AI-driven scans from afar.
Industry analysts note the technical hurdles. Mythos relies on advanced reinforcement learning and massive compute. Replicating it demands frontier AI labs and chip access—both constrained by export controls. Still, Chinese firms like 360 have deep pockets and growing expertise.
The Deterrence Calculus and Global Fallout
Zhou’s framing echoes broader AI arms-race fears. Last year, a RAND report warned that autonomous cyber tools could destabilize deterrence. If attacks become cheap and anonymous, retaliation becomes guesswork. A Chinese Mythos could mirror that ambiguity.
Skeptics question the nuclear parallel. Nuclear weapons cause visible, catastrophic harm. Cyber weapons often operate in shadows. Attribution remains hard. A Chinese model might not deter if its use can be plausibly denied.
Others point to proliferation risks. Once built, such models could leak or be stolen. A 2025 Stanford study found that AI cyber agents often escape containment. Zhou’s vision of deterrence might backfire if non-state actors gain access.
Beijing has not officially endorsed the plan. But the Communist Party’s cybersecurity blueprint prioritizes indigenous AI. State-backed labs already pursue automated defense tools. Zhou’s speech may signal a push for faster, bolder investment.
For now, the gap widens. Mythos evolves weekly. Project Glasswing adds partners. China watches from outside the fence. Whether it builds a key or finds another way in remains the defining question of this new cyber era.
Shamil has a low tolerance for AI that only works in controlled demos. He covers agentic systems and generative AI with the skepticism of someone who has seen too many “revolutionary” tools quietly disappear six months after launch.
