AI Governance, Military Contracts, and Strategic Risk
Stratechery founder Ben Thompson analyzes the Anthropic-Department of War conflict, arguing that AI's power dynamics necessitate a shift from voluntary alignment to state-mandated governance. The analysis highlights the economic unsustainability of government-only AI development and the geopolitical trade-offs of chip export controls.
The End of Voluntary AI Alignment
The recent designation of Anthropic as a supply chain risk by the Department of War marks a pivotal shift in the relationship between frontier AI developers and the state. Ben Thompson’s analysis argues that the era of voluntary alignment is ending. As AI capabilities approach those of nuclear weapons in terms of power dynamics, the assumption that private companies can unilaterally impose safety restrictions is no longer tenable. The state, possessing the monopoly on violence, will inevitably seek to control or compel access to these technologies, rendering corporate ethics insufficient as a regulatory framework.
Economic Realities of Frontier AI
A critical economic insight is that government-only funding for frontier AI is unsustainable. The capital expenditure required for training state-of-the-art models, approaching hundreds of billions of dollars annually, necessitates a broad commercial customer base. This economic reality forces AI companies to serve the general market, which in turn introduces complex dynamics where the government is a small but powerful customer with the ability to disrupt the entire business model. This creates a structural tension: companies need government contracts for legitimacy and security clearance, but resisting government mandates risks existential threats to their commercial viability.
Geopolitical Trade-offs and Chip Controls
The debate extends to geopolitical strategy, particularly regarding chip export controls. Thompson argues that absolute restrictions on selling advanced chips to China may increase the likelihood of military conflict, such as a strike on Taiwan. He posits that a safer equilibrium might involve allowing limited Chinese access to maintain dependency, rather than attempting total decoupling which could provoke aggressive military action. This highlights the need for nuanced trade-off analysis in tech policy, where absolute positions often lead to worse outcomes.
Strategic Recommendations for Tech Leaders
For tech leaders, the path forward involves recognizing that AI is a source of power that will be regulated as such. Companies must move beyond internal ethics boards and engage in the political process, potentially through Political Action Committees (PACs), to lobby for specific legislation addressing digital surveillance and AI governance. The goal is to replace ad-hoc executive decisions with clear legal frameworks, ensuring that safety standards are codified in law rather than left to the discretion of private executives or the whims of political administrations. This approach mitigates the risk of adversarial government action and aligns corporate strategy with broader democratic processes.
Key insights
-
AI is transitioning from a commercial product to a source of state power, necessitating a shift from voluntary corporate alignment to state-mandated governance. The analogy to nuclear weapons proliferation is apt, as the state will seek to control or compel access to prevent asymmetric power advantages.
Impact: Tech companies must prepare for increased regulatory scrutiny and potential government mandates, requiring robust legal and political strategies to navigate this new landscape.
-
The economic scale of frontier AI development makes government-only funding unsustainable. Companies must serve broad commercial markets to achieve the necessary scale, which introduces complex dynamics where the government is a small but powerful customer with the ability to disrupt the entire business model.
Impact: This economic reality forces AI companies to balance commercial growth with government relations, creating a structural tension that requires careful management to avoid existential threats.
-
Absolute restrictions on chip exports to China may increase geopolitical instability by provoking military conflict over Taiwan. A safer equilibrium might involve allowing limited Chinese access to maintain dependency, rather than total decoupling which could lead to aggressive military action.
Impact: Policy makers and tech leaders must consider the unintended consequences of export controls, recognizing that total decoupling may be less stable than managed dependency.
-
Existing laws assume physical friction for surveillance, which AI and computing have obviated. This creates massive loopholes that allow for mass domestic surveillance, requiring new legislation to address these digital-specific issues.
Impact: Companies must lobby for specific digital surveillance acts to close these legal gaps, ensuring that safety standards are codified in law rather than left to the discretion of private executives.
-
Arguing that AI models are not yet capable for military use is a weak defense against government mandates. The state’s primary concern is control and precedent, not current technical limitations, making capability arguments ineffective in high-stakes negotiations.
Impact: Tech companies should focus on legal and political strategies rather than technical limitations when negotiating with the government, as the latter are unlikely to sway state decisions on power dynamics.
Action items
-
Establish a dedicated political strategy team to engage with legislators and form Political Action Committees (PACs) to lobby for specific AI governance legislation. This approach aligns corporate interests with democratic processes and reduces adversarial risk.
Impact: Proactive political engagement can help shape regulatory frameworks in a way that protects corporate interests while addressing public concerns, reducing the likelihood of adversarial government action.
-
Develop a comprehensive risk assessment for government contract dependencies, identifying potential vulnerabilities in the event of political shifts or regulatory changes. This includes modeling scenarios where the government mandates access or restricts commercial operations.
Impact: Understanding and mitigating these risks can help companies maintain operational stability and protect their commercial viability in the face of increasing state involvement in AI.
-
Re-evaluate chip export control strategies by considering the geopolitical implications of total decoupling versus managed dependency. Engage with policy makers to explore nuanced approaches that reduce the risk of military conflict while maintaining national security.
Impact: A more nuanced approach to export controls can help stabilize geopolitical relations and reduce the risk of conflict, creating a more predictable environment for business operations.
-
Invest in legal infrastructure to address digital surveillance loopholes, working with legal experts to draft and advocate for new legislation that closes these gaps. This includes collaborating with civil liberties organizations to build a broad coalition for support.
Impact: Closing legal loopholes can enhance public trust in AI companies and reduce the risk of regulatory backlash, while also providing a clearer framework for responsible AI development.
-
Shift internal communication strategies to emphasize the economic and strategic necessity of broad commercial markets for AI development, rather than relying on technical capability arguments. This helps align internal stakeholders and external partners on the core business case.
Impact: Clear communication of the economic realities of AI development can help build support for commercial strategies and reduce internal resistance to government engagement, fostering a more unified approach to navigating the new landscape.
Quotes
“You may not be interested in politics, but politics has an interest in you.”
“If AI is as powerful as its builders claim, the people with guns are going to want to say.”
“The government is a small customer. The government is also the entity, again, not to be but with guns.”