US, Chinese AI Experts Urge Cooperation on Cyber Risks as Open Models Near Mythos-Level Power
Updated
Updated · WIRED · Jun 24
US, Chinese AI Experts Urge Cooperation on Cyber Risks as Open Models Near Mythos-Level Power
3 articles · Updated · WIRED · Jun 24
Summary
US and Chinese researchers at a Beijing AI conference warned that increasingly agentic models could soon enable cyberattacks and broader systemic failures unless the two countries cooperate on safety.
Sessions highlighted concrete dangers in AI-generated code, agentic tool use and automated social engineering, while MIT's Stephen Casper argued shared work on AI risks would outweigh national-security costs.
Open-weight models are sharpening the dilemma: Chinese systems such as Z.ai's GLM 5.2 already show frontier coding and agentic abilities, and 360 Security Technologies said this week its hacking model matches Anthropic's Mythos.
That risk is already affecting access and release decisions, with Washington tightening model restrictions and a source saying some advanced Chinese models are no longer being released as open source for security reasons.
Experts said the longer-term goal is shared safety principles and technical standards that reduce systemic risk without exposing sensitive operational details.
As the US-China AI race accelerates, can safety talks prevent a 'Chernobyl moment' before it is too late?
As powerful open-weight AI spreads globally, what stops it from becoming the next uncontrollable weapon for rogue actors?
Claude Mythos Unleashed: The 2026 AI Breakthrough That Upended Cybersecurity, Geopolitics, and Global Governance
Overview
In April 2026, Anthropic's Claude Mythos marked a turning point in global technology by demonstrating unprecedented AI capabilities. This 'Mythos Moment' immediately alarmed governments and financial institutions, exposing systemic vulnerabilities and forcing a policy wake-up call. Mythos-class models advanced cybersecurity by helping identify and respond to sophisticated threats, while Anthropic's Project Glasswing gave vetted companies early access to these powerful tools. However, the emergence of such advanced AI also raised urgent concerns about misuse, governance, and the need for international cooperation, as the rapid pace of development outstripped existing security and policy frameworks.