The world’s two most powerful AI players gathered in Washington for a three-day summit that placed artificial-intelligence risk at the centre of diplomatic talks. President Donald Trump welcomed Chinese President Xi Jinping amid an atmosphere of heightened anxiety: CEOs, former tech insiders and policy-makers have been sounding Skynet-style warnings on social platforms, and the stakes now feel unmistakably political.
On the same day the United Nations opened its annual General Assembly, the heads of leading AI firms appealed to the global body for a firm policing mandate. Sam Altman chief executive of OpenAI addressed the UN Security Council with a stark warning: “We could lose control of the future to AI.” His comment underscored how quickly the conversation has shifted from speculative tech talk to a frontline issue of international security.
What the current AI safety patchwork looks like
Over the past four years, a collection of national and multilateral initiatives has emerged to tame the accelerating power of generative models. In Washington, Congress and the White House have convened hearings, commissioned expert panels and drafted draft regulations. Parallel efforts exist at the United Nations where scientific panels release yearly assessments of the most advanced systems, and at the G7 which runs a reporting mechanism that obliges participating AI firms to disclose model-building practices and risk-mitigation steps.
Several countries have created government-funded AI safety institutes tasked with stress-testing new products before they reach the market. Meanwhile, private companies have signed voluntary pledges to curb election interference, block deep-fake proliferation, and embed safety checks into development pipelines. Together these measures form a loosely-woven safety net, but they lack a single, rapid-response conduit that can be activated when an incident threatens multiple jurisdictions.
US-China dialogue: a possible shortcut
The Washington summit offered a concrete opportunity to stitch together a more coherent emergency framework. U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng were reported to be discussing a dedicated hotline for AI-related security incidents. Such a line would operate much like the historic nuclear-crisis hotlines between superpowers: a direct, technical channel aimed at de-escalating misunderstandings, coordinating responses to model-driven cyber-attacks, and ensuring swift compliance with agreed safety norms.
Chinese officials have so far resisted broader U.S. calls for universal AI rules, pointing out that domestic legislation in China already imposes stringent oversight. Nonetheless, the prospect of a bilateral rapid-response link signals a willingness to move past rhetoric and create a functional, if narrow, safeguard. Whether the three-day summit will cement that link remains uncertain, but the dialogue itself marks a shift from isolated national policies toward a shared crisis-management mindset.
Blueprint for a 90-day interim response system
Experts suggest that, while a comprehensive global treaty may still be years away, a stop-gap mechanism can be built within 90 days. The proposal leverages existing structures and focuses on three practical pillars:
Incident-reporting protocol
When a serious AI malfunction occurs, participating firms would file a confidential report within 72 hours to a designated national responder and a small, technical secretariat. The process could expand on the OECD’s AI Incidents Monitor, triggering a multi-stakeholder investigation and the eventual release of an anonymised lessons-learned brief.
Pre-release safety declaration
Before launching a new model or major upgrade, developers would publish a standardized declaration covering safeguards, residual risks, escalation thresholds and third-party audit results. Built on the G7 Hiroshima AI Process framework, this would function like the post-2008 financial-crisis stress tests, allowing regulators and the public to compare safety postures without picking a single “best” company.
Limited-scope AI hotline
Initially limited to the United States and China, a technical hotline would enable immediate coordination on acute AI threats. The channel’s remit would stay narrowly defined—preventing dangerous misunderstandings, responding to AI-enabled cyber incidents, and verifying compliance with the newly-adopted reporting standards. Over time, other nations could be invited to join, gradually expanding the safety net.
Critics note that the plan leans heavily on Western institutions and the U.S.–China relationship, potentially sidelining the Global South. Yet the alternative—waiting for a universal treaty while AI models evolve at breakneck speed—poses a risk many deem unacceptable. The urgency expressed by OpenAI’s Sam Altman at the UN highlights the narrow window for decisive action.
In the months ahead, the combination of national safety institutes, G7 reporting, UN scientific assessments, and a fast-track 90-day response framework could provide the first coordinated line of defence against a technology that is already outpacing regulation. As the summit concludes and the world watches, the question shifts from “Can we control AI?” to “How quickly can we react when it slips beyond our grasp.”
