"Clawed" is an essay by Dean W. Ball, published in his newsletter Hyperdimensional on 2026-03-02, analyzing the conflict between Anthropic and the Department of War (DoW, formerly the Department of Defense) over contractual use restrictions on Claude in classified military contexts. Ball treats the episode as a case study in how democratic institutions are handling frontier AI, framing it as a "death rattle" of republican governance. The essay argues that the proper locus of control over frontier AI is the central policy question of the era and that, on this first major test, governmental institutions performed poorly.
Background
Ball recounts the deal's history as follows. During the Biden administration, Anthropic negotiated terms for Claude's use in classified contexts subject to two restrictions: no mass surveillance of Americans, and no use to control lethal autonomous weapons. The Trump administration expanded the deal in July 2025, accepting the same terms. Trump officials subsequently reversed their position. By Ball's account, the reversal was not driven by an imminent desire to conduct surveillance or field autonomous weapons, but by an objection to privately imposed policy limitations on the military.
The DoW then threatened to designate Anthropic a "supply chain risk," a power Ball describes as normally reserved for firms controlled by foreign adversaries, such as Huawei. Secretary Hegseth threatened to prevent all military contractors from holding "any commercial relations" with Anthropic, a step Ball argues would effectively cut off the company's cloud computing and chip purchases and force investors including Nvidia, Google, and Amazon to divest.
Summary of argument
Ball advances several claims about the episode:
- Contractual use restrictions are ordinary. Thousands of defense contractors impose them, and Anthropic's terms resemble standard practice, differing in that they apply "policy constraints" rather than "technical constraints."
- Both the Biden and Trump administrations accepted the terms, so the arrangement was not a departure from defense-contracting norms.
- A proportionate response would have been narrow: cancel the contract and issue DFARS guidance on AI use-restriction terms, what Ball describes as the "least restrictive means" approach.
- Instead the DoW chose what Ball calls corporate punishment, combining a supply-chain-risk designation with a blanket commercial ban. In Ball's words, "the United States Secretary of War announced his intention to commit corporate murder."
- The episode produces collateral damage: every software vendor now faces elevated regulatory risk; Chinese AI such as DeepSeek becomes somewhat less risky relative to American AI from a regulatory standpoint; and the American AI Exports Program is, in Ball's phrase, "dead on arrival."
Ball concludes that "the limitations imposed in the law on what we know today as 'the government's' use of AI will be of paramount importance," and cautions against assuming that governmental control of AI is synonymous with democratic control.
Relation to DoD Directive 3000.09
The conflict described by Ball operates above the policy baseline set by DoD Directive 3000.09, "Autonomy in Weapon Systems," reissued January 25, 2023 by Deputy Secretary Hicks. The directive is the US government's governing policy on autonomous and semi-autonomous weapons. It regulates lethal autonomous weapon systems (LAWS) by requiring senior review by USD(P), USD(R&E), and the VCJCS before development and fielding, but does not prohibit them.
Anthropic's contractual term that Claude "cannot be used to control lethal autonomous weapons" is therefore stricter than Directive 3000.09 requires, functioning as a private-sector policy floor layered above the institutional one. The administration's objection, per Ball, is not that Anthropic violated the directive but that Anthropic imposed a policy ceiling the directive would otherwise allow the DoW to push up against. Because Directive 3000.09 already provides a policy instrument through which the US could set or change its position on LAWS use of Claude, Ball's "policy constraint via contract vehicle" critique turns on the observation that the DoW bypassed that instrument and resorted to supply-chain-risk coercion instead.
Relation to the ICRC position
Ball's essay operates in a republican-governance register, concerned with who should set AI use policy, through what institutional vehicle, and what democratic theory licenses the answer; it is largely agnostic on the substantive content of AI use restrictions.
The ICRC's 2021 position on autonomous weapon systems, reinforced in its 2024 UN Secretary-General submission and President Mirjana Spoljaric's 3 May 2024 Vienna Conference statement, operates in a complementary international humanitarian law register. Where Ball asks who sets the policy, the ICRC addresses what substantive limits that policy must respect: categorical prohibitions on unpredictable AWS and anti-personnel AWS, plus strict regulation of remaining AWS to preserve meaningful human control.
Read alongside the ICRC position, Anthropic's contractual exclusion of Claude from lethal autonomous weapons aligns substantively with the ICRC's call for prohibition of anti-personnel AWS, while operating at the private-contract level that Ball argues is the wrong vehicle. The two critiques are independent: a procedural one (Ball's, that corporate coercion was the wrong remedy) and a substantive one (the ICRC's, concerning the content of the restriction), and the DoW's objection runs against both.
Provenance
The essay is an opinion piece presenting Ball's argument and framing; its factual recitation of the deal's terms and the DoW's threatened actions is attributed to Ball and not independently corroborated on this page. Confidence is medium pending further sourcing.
Relationships
- related: Anthropic — the company whose conflict with the US government over AI use in military contexts the essay documents
- related: DoD Directive 3000.09 — Autonomy in Weapon Systems (source summary) — the institutional US policy instrument the conflict operates against and above
- related: ICRC Position on Autonomous Weapon Systems (source summary) — IHL-side substantive counterpoint to both DoD 3000.09 and the DoW's position; Anthropic's contractual LAWS exclusion aligns with the ICRC's call for prohibition of anti-personnel AWS
- related: International Committee of the Red Cross (ICRC) — the international humanitarian law authority whose position provides the substantive register complementing Ball's governance critique
- related: AI and Authoritarianism — the domestic dimension of a democratic government demanding unrestricted AI access for surveillance and autonomous weapons
- related: Export Controls (AI) — Ball argues the supply-chain-risk designation makes American AI riskier to use globally, undermining the export strategy
- related: America's AI Action Plan — Ball argues the Anthropic-DoW conflict contradicts the Plan's stated vision of private-sector-led AI innovation
- related: EO 14365 — another instance of the Trump administration using executive power to pressure private AI companies