Updated
Updated · WIRED · Aug 27
US, China Researchers Explore AI Safety Cooperation as Hacking Risks Rise With 2 AI Powers
Updated
Updated · WIRED · Aug 27

US, China Researchers Explore AI Safety Cooperation as Hacking Risks Rise With 2 AI Powers

3 articles · Updated · WIRED · Aug 27

Summary

  • Researchers in the US and China are pushing for AI safety cooperation as more capable AI agents raise fears of hacking sprees, self-replication and other cross-border failures.
  • Cybersecurity has become the main driver: experts in Beijing and Shanghai said agentic safety is now a major focus, with concern centered on models that could attack networks, evade controls or trigger wider systemic incidents.
  • Possible cooperation ranges from shared technical benchmarks to crisis communication channels, so Washington and Beijing could quickly signal when an aggressive AI action is a mistake rather than an attack.
  • Those efforts still face export controls, mistrust and limits on direct collaboration, even as Chinese researchers say they want US companies and academics involved in testing model hacking capabilities.
  • The debate cuts against the usual zero-sum AI race narrative, with researchers in both countries warning that a financial flash crash or AI-driven cyberattack could become a 'Chernobyl moment' neither side can contain alone.

Insights

Are new international AI safety benchmarks a genuine shield against rogue agents or a hidden weapon to monopolize the tech race?
With AI supply chains deeply intertwined, will geopolitical export controls accidentally trigger the autonomous cyberattacks they aim to prevent?
Could a self-propagating AI worm force rival nations to share their deepest cybersecurity secrets before a global network collapse?