The coming US-China rapprochement

Nonzero  |  Robert Wright

United States artificial intelligence policy analyst Jeremie Harris recently advocated for an enforceable and verifiable AI treaty with China to mitigate catastrophic cybersecurity risks. This urgent proposal emerges as both superpowers race to deploy advanced models like Anthropic's Mythos, which remain highly vulnerable to jailbreaking by hostile state-sponsored cyber actors.

The Trump administration previously paused the release of Anthropic's Fable and OpenAI's ChatGPT-5.6 to assess potential national security threats. However, unilateral American regulatory pauses risk ceding technological leadership to Beijing unless formal bilateral guardrails are established. While some hawkish analysts suggest sub-threshold cyber warfare to disrupt Chinese domestic development, a formal bilateral dialogue established during the Beijing summit offers a more stable mechanism to manage these non-zero-sum existential threats. Sustaining this nascent diplomatic channel could eventually foster organic transparency and broader trans-Pacific scientific collaboration, paving the way for a wider geopolitical rapprochement despite deep-seated military rivalries.

Comment

Verification of software-based capabilities remains a fundamental vulnerability in modern arms control. The 2015 Obama-Xi cyber agreement failed due to the invisible nature of digital espionage. National technical means of verification cannot easily detect hidden neural network weights or dual-use algorithmic code. Beijing will likely exploit any regulatory pause to advance its own offensive cyber warfare programmes.

To read in details click here