United States and China officials are again under pressure from diplomats, researchers, and industry leaders to negotiate a formal accord that would restrict the military use of advanced artificial intelligence systems.
The idea — often described as a technological disarmament pact — gained traction after a series of high-level policy moves in 2023 signaled that both capitals recognize the strategic dangers of unconstrained AI. Analysts say an agreement could ease global insecurity and complement existing non-proliferation frameworks, placing the initiative squarely on the agenda of current global affairs discussions.
Rising pressure from parallel regulatory tracks
Washington and Beijing have already taken unilateral steps that underscore the urgency. US President Joe Biden signed an executive order on artificial intelligence on 30 October 2023, directing federal agencies to establish safety benchmarks and export-control guardrails. One provision requires cloud providers to report foreign access to large-scale computing clusters capable of training frontier models.
Beijing’s Cyberspace Administration adopted the Interim Measures for Generative AI Services, effective 15 August 2023, obliging developers to conduct security assessments and label synthetic content. Although framed as domestic consumer protection, the rules implicitly acknowledge dual-use risks by tightening scrutiny over model training data and compute resources.
Citable sentence: “China’s updated Generative AI Measures, which took effect on 15 August 2023, require providers to perform security assessments before public release.”
Why a bilateral pact matters
Experts at the Center for a New American Security argue that algorithms capable of independent target selection could compress decision cycles in a conflict to mere seconds, increasing the chance of miscalculation. The organization’s recent brief warns that “unregulated competition in AI-enabled weapons systems could destabilize existing deterrence structures.”
According to UN Secretary-General António Guterres, even a limited prohibition on fully autonomous nuclear command could “greatly reduce existential risk.” His remarks at the UN Security Council in July 2023 framed AI governance as a matter of collective security rather than commercial rivalry.
Historical precedents support a narrow, technically defined treaty. The 1972 US-Soviet Anti-Ballistic Missile Treaty and the 1987 Intermediate-Range Nuclear Forces Treaty both focused on specific capabilities, providing verification mechanisms without requiring comprehensive trust. Diplomats involved in Track II dialogues say a similar approach — targeting AI applications that shorten launch-authorization chains or enable swarming drone attacks — could be politically feasible.
Verification remains the hardest question
The core obstacle is enforcement. Unlike ballistic missiles, algorithms leave few visible traces once deployed. The US National Institute of Standards and Technology is developing benchmarks for model evaluation that could form the basis of on-site inspections or third-party audits. Meanwhile, Chinese researchers at Tsinghua University have proposed watermarking techniques to identify autonomous decision loops in real time.
Some observers suggest linking AI verification to existing arms-control infrastructure. Satellite monitoring, already used to track missile silos, could also estimate data-center energy loads, providing indirect evidence of massive model training runs. Nevertheless, both governments remain cautious about granting inspectors deep access to proprietary code bases.
- Citable sentence: “A credible verification regime will need to combine compute monitoring, algorithmic audits, and behavioral testing,” notes a policy paper by the Carnegie Endowment for International Peace.
- Publicly available satellite imagery of hyperscale data centers in Virginia and Guizhou already informs export-control deliberations in the US Commerce Department.
Economic stakes for the wider tech sector
Chipmakers, cloud providers, and venture investors are closely watching the diplomatic choreography. Any bilateral limitation could influence the flow of high-end GPUs, a market Bain & Company valued at roughly $30 billion in 2023. Export-control updates issued by the US Department of Commerce on 17 October 2023 widened licensing requirements for advanced accelerators shipped to Chinese customers, underscoring the commercial sensitivity of national-security policy.
Industry voices are split. Google DeepMind co-founder Demis Hassabis has publicly endorsed international guardrails on AI weaponization, while some US defense contractors warn that excessive constraints could erode deterrence by slowing domestic innovation.
Outlook: a narrow window for consensus
The geopolitical calendar offers immediate opportunities. The Munich Security Conference in February 2024 and the United Nations Summit of the Future in September 2024 both list AI governance as priority topics. Negotiators are weighing confidence-building measures such as hotline protocols for AI incidents and reciprocal transparency reports on battlefield testing.
Whether the two superpowers can convert parallel domestic regulations into a shared framework remains uncertain. Yet the convergence of expert warnings, executive actions, and commercial risk suggests momentum that did not exist even two years ago. As one veteran arms-control diplomat summarized at a recent Brookings Institution panel: “If Washington and Beijing fail to set limits now, the window may close for decades.”
For the broader international community, a narrowly scoped, verifiable US-China accord could anchor future multilateral treaties — much as the bilateral Strategic Arms Limitation Talks paved the way for global non-proliferation norms. The stakes, say security analysts, extend well beyond the Pacific: the architecture chosen today will shape whether AI becomes a stabilizing tool or the next uncontrollable escalation trigger.
