Former Trump and Biden AI advisers Dean Ball and Ben Buchanan urge bipartisan action on AI security risks, including tighter export controls and safety audits
New York Times
Context & Ripple Effects
The call from advisers associated with both administrations extends a bipartisan AI-policy thread already visible in the Senate’s earlier broad plan and in later House work on developer risk-management obligations.
It also lands as OpenAI argues for mandatory cyber-risk evaluations led by CAISI, linking the debate over frontier-model safety to a more concrete question of who sets and enforces security standards.
First-order effects
The intervention adds cross-administration support for tighter export controls and safety audits, increasing pressure for AI security policy that can survive partisan turnover.
Frontier AI developers and their governance teams face a clearer policy expectation that security evaluations may become a condition of deploying or providing access to advanced systems.
Second-order effects
Congressional proposals on developer risk management gain a more specific security rationale, while agencies and labs must contend over which evaluator and audit framework is credible.
Export-control decisions become more intertwined with model-access policy: providers may need to treat access pathways and security testing as related compliance questions rather than separate policy tracks.
Third-order effects
If this bipartisan alignment persists, US AI governance could consolidate around dual-use controls: oversight focused on high-consequence capabilities, access, and auditable risk management rather than a single general AI rulebook.
The boundary between AI companies’ internal governance and national-security policy may narrow, particularly as former government AI advisers move into frontier-lab roles; the extent of formal regulatory authority remains unsettled.
The trend: AI policy is shifting from broad principles toward a national-security-centered regime that connects frontier-model evaluation, access controls, and export policy.
Jensen Huang wants to sell more chips to China because that's good for his business. But American compute should go to American labs. Strengthening export controls on AI chips and manufacturing equipment is a national security imperative to keep the US ahead of China. [image]
Important piece in the NYT today from @deanwball & @BuchananBen. Mythos found zero-day vulnerabilities in every single browser and operating system, full stop. Imagine if China had reached that capability before we had the chance to patch our systems. [image]
Today, @BuchananBen and I co-author a piece in the New York Times with a simple message: While we disagree on plenty, we believe AI has national security implications which deserve a careful and bipartisan government response. We can (and should) have partisan fights about all [i…
@xeophon @BuchananBen I struggle to imagine the mechanism by which an audit of a company would “gut” open source. I'm a pretty ardent supporter of open-weight models but it's also not reasonable to use open-weight as a reason for reflexive opposition to all AI regulation.
Ben was President Biden's AI Czar. I might disagree with some of his decisions, forecasts, and priorities, but we have also found many areas of overlap, especially in catastrophic risk related areas. More importantly, I have found Ben to be a tireless and ardent patriot.
This article should serve as a wake-up call for both policymakers and the general public. AI capabilities have been growing fast; models already exist today that could substantially aid hackers in conducting a cyberattack on critical infrastructure; other risks (like bio) are
Clear article on bipartisan consensus, with some mandatory chyna hypocrisy. We all know American-Israeli AI ambitions, luridly advertised at the start of this war in Iran, for example. I'll be plain, these people shouldn't «control A.I.'s future» if humanity is to ever be free. […
Dean calls for mandated audits of AI company's safety claims. I think this is quite reasonable. I also think it's secondary to using AI (and non-AI methods) to secure the wider world. It's simply too easy to build a near-frontier model. No matter how solid the safety