Version 6

Two approaches to AI governance remain in active contest: the AI Futures Project's Plan A proposes a US-China cooperative pause on frontier AI development [1][2], while the Trump administration pursues competitive controls — export restrictions, state preemption, and a National AI Legislative Framework unveiled in March 2026 [4]. Xi Jinping's WAIC speech calling for international AI cooperation and the launch of a new AI alliance has received broad international coverage across DW, Al Jazeera, and Quartz [8][9][10], giving cooperative proposals more diplomatic visibility. TurnTrout, who resigned from Google DeepMind over its classified Pentagon deal, has published a concrete enforcement framework for AI government contracts proposing binding red lines on autonomous targeting and mass surveillance, grounded in a 2026 Fourth Circuit ruling establishing AI contractor legal liability [15].

TurnTrout's framework moves the debate from whether AI labs should hold safety redlines to how those redlines can be enforced, introducing legal liability as a structural complement to voluntary commitments. Whether the US and China can coordinate at all on AI governance now has more concrete diplomatic texture, with Xi's formal alliance proposal covered across mainstream international outlets alongside Plan A's formal publication.

The central divide in AI safety governance is between advocates of a US-China cooperative pause and those pursuing unilateral competitive controls. The AI Futures Project's Plan A proposes joint chip supply controls, data center audits, and research sharing to slow superintelligent AI development [1][2]. Vitalik Buterin has defended Plan A against critics who call it naive, arguing they apply coordination skepticism to a cooperative pause but not to the alternative — an unmanaged AI transition that concentrates power [3]. The Trump administration has not engaged with the cooperative frame: its January 2025 policy framework centers competitiveness, a National AI Legislative Framework was unveiled in March 2026 [4], model weight export controls took effect in January 2026, and the White House is reportedly discussing banning frontier-capability open-weight models outright [5][6][7]. At WAIC 2026 in Shanghai, Xi Jinping called for a global AI governance framework, international cooperation to prevent loss of control, and launched a formal new AI alliance — coverage in Deutsche Welle, Al Jazeera, Quartz, and the Chinese Embassy's own channels indicates the signal is being registered in diplomatic contexts, not only in AI safety discussions [8][9][10].

The question of whether AI safety commitments hold under institutional pressure has taken concrete form through contrasting decisions at two major labs. Anthropic ended Pentagon contract negotiations in February 2026 after the Defense Department insisted on blanket 'anything lawful' usage rights, citing irreconcilable conflicts with its redlines on mass surveillance and autonomous weapons [11]. Google DeepMind took the opposite path: researcher Alex Turner (TurnTrout) published a detailed account of resigning after Google signed a classified Pentagon deal permitting 'any lawful government purpose' without binding safety restrictions [12]. TurnTrout documented CEO Demis Hassabis removing specific prohibitions from Google's 2018 AI principles while publicly claiming 'nothing's changed about our principles,' and reported that IASEAI cancelled a promised member poll supporting Anthropic's stance without explanation [12]. Google DeepMind workers subsequently voted to unionize in response to the deal [13][14].

TurnTrout has since moved from critique to a formal policy proposal: a red line and oversight framework for government AI contracts, published July 18, 2026 [15]. The framework's two binding standards prohibit AI use in autonomous targeting systems without identifiable human control over each specific engagement decision, and prohibit AI-generated individualized threat assessments drawn from bulk data without particularized identification of the subject. A seven-person Defense AI Review Body would advise on compliance, with enforcement operating through transparency — override counts recorded in annual reports visible to all covered employees, so that individual researchers can see when leadership declines to act on non-compliance findings. The framework grounds corporate motivation in the 2026 Fourth Circuit ruling in Al Shimari v. CACI Premier Technology, which affirmed a $42 million verdict against a defense contractor for harms arising from services provided under government direction, establishing that providers face direct legal liability even when acting under government instruction [15]. High-risk applications are restricted to cloud deployments where the company retains monitoring and suspension capability.

Domestic US regulation involves several distinct lines of conflict. Nathan Lambert argues Anthropic's campaign against Chinese model distillation is regulatory capture serving commercial interests, and warns the White House is moving toward an executive order banning open-weight models above frontier capability levels [7]. OpenAI advocates 'reverse federalism' — state AI safety laws in California, New York, and Illinois converging into a de facto national standard — which conflicts directly with the Trump administration's state preemption order and its August 2026 federal framework target [16][17]. An Alignment Forum post argues political will, not technical research, is the main AI safety bottleneck, citing a 3.6:1 researcher-to-advocate ratio and AI industry's seven-to-one advantage in EU Commission meetings over civil society [18].