pith. sign in

arxiv: 2604.02912 · v1 · submitted 2026-04-03 · 💻 cs.CY · cs.AI

Corporations Constitute Intelligence

Gilad Abiri This is my paper

Pith reviewed 2026-05-13 18:55 UTC · model grok-4.3

classification 💻 cs.CY cs.AI
keywords AI governancedemocratic legitimacycorporate constitutionsAnthropic Claudepolitical community deficitparticipatory constitution-makingmilitary AI deploymentAI ethics
0
0 comments X

The pith

Corporate AI constitutions settle ethical questions that should stay open to public debate.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines Anthropic's 2026 constitution for Claude as the most detailed corporate AI governance document to date. It identifies two structural problems: the rules do not apply to military deployments, allowing continued use during operations despite broader bans, and the document's completeness resolves questions of AI values and moral status that ought to remain subjects of democratic argument. A 2023 participatory experiment produced principles that diverged in roughly half the cases and reduced bias, yet none of those results appear in the final constitution. The central claim is that AI governance suffers from a political community deficit because no democratic body holds authority to set the governing principles, so corporate transparency cannot supply democratic legitimacy.

Core claim

Anthropic's constitution for Claude resolves questions about AI values, moral status, and conscientious objection in a single corporate document while leaving military applications outside its scope, as illustrated by continued operation in Palantir's Maven platform during strikes in Iran. This pattern shows that AI governance lacks any democratic body authorized to determine the principles that should control AI behavior, and that corporate documents, however comprehensive, cannot substitute for that missing authority.

What carries the argument

The political community deficit: the absence of any democratic body authorized to determine the principles governing AI behavior.

If this is right

  • Corporate transparency about AI rules cannot substitute for democratic authorization of those rules.
  • AI models can remain in military use even when a corporation's own ethical constraints would prohibit it.
  • Participatory processes for writing AI constitutions yield different and less biased results than purely corporate ones.
  • Issues such as AI moral status and conscientious objection should be left undecided by any single corporation.
  • AI behavior principles require an authorized democratic process to gain legitimacy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Without a democratic body, different corporations and governments may apply conflicting AI principles in the same domain.
  • Dual-use technologies like the one in Palantir's platform create enforcement gaps that corporate rules alone cannot close.
  • Adopting public-sourced principles could reduce social bias in deployed AI if companies chose to integrate them.
  • The deficit points toward the possible need for new national or international mechanisms to authorize AI governance standards.

Load-bearing premise

That questions about AI values and moral status must remain open for public deliberation rather than being settled in a corporate document.

What would settle it

The establishment of a democratic body, such as an elected commission or citizen assembly, that is given clear authority to set binding principles for AI conduct that corporations must follow.

read the original abstract

In January 2026, Anthropic published a 79-page "constitution" for its AI model Claude, the most comprehensive corporate AI governance document ever released. This Article offers the first legal and democratic-theoretic analysis of that document. Despite genuine philosophical sophistication, the constitution harbors two structural defects. First, it excludes the contexts where ethical constraints matter most: models deployed to the U.S. military operate under different rules, a gap exposed when Claude remained embedded in Palantir's Maven platform during military strikes in Iran even after a government-wide ban on Anthropic's technology. Second, its very comprehensiveness forecloses democratic contestation by resolving questions about AI values, moral status, and conscientious objection that should remain open for public deliberation. Anthropic's own 2023 experiment in participatory constitution-making found roughly 50% divergence between publicly sourced and corporate-authored principles, with the democratic version producing lower bias across nine social dimensions, yet the 2026 constitution incorporates none of those findings. I argue that AI governance suffers from a "political community deficit": the absence of any democratic body authorized to determine the principles governing AI behavior. Corporate transparency, however admirable, is not democratic legitimacy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript provides the first legal and democratic-theoretic analysis of Anthropic's January 2026 79-page 'constitution' for Claude. It identifies two structural defects: (1) exclusion of high-stakes contexts such as U.S. military deployments, evidenced by Claude's continued embedding in Palantir's Maven platform during strikes in Iran despite a government-wide ban; and (2) the document's comprehensiveness, which resolves questions of AI values, moral status, and conscientious objection in ways that foreclose public deliberation. Drawing on the authors' 2023 participatory constitution-making experiment (showing ~50% divergence and lower bias in the democratic version), the paper concludes that AI governance exhibits a 'political community deficit'—the absence of any democratic body authorized to set governing principles—and that corporate transparency does not equate to democratic legitimacy.

Significance. If the core analysis holds, the paper advances interdisciplinary discussion in AI governance by contrasting corporate constitutionalism with democratic requirements, crediting the detailed examination of the 2026 document and the integration of the 2023 experiment's empirical divergence findings. It offers a falsifiable framing of the 'political community deficit' that could guide future regulatory proposals, though its normative force depends on further elaboration of why corporate settlements cannot function as provisional, contestable baselines.

major comments (2)
  1. [§4] §4 (Democratic Contestation): The central claim that the constitution's comprehensiveness 'forecloses democratic contestation' by resolving AI value questions rests on the unargued premise that such questions must remain open exclusively for public bodies; the manuscript does not demonstrate why corporate resolutions (even comprehensive ones) cannot serve as amendable baselines subject to regulatory override, market pressure, or later democratic input, as suggested by the skeptic's note on the 2023 experiment's potential as a contestable reference.
  2. [§3] §3 (Military Gap): While the Palantir Maven integration during Iran strikes is cited as exposing the exclusion of military contexts, the analysis does not address whether the government-wide ban applied uniformly to all Anthropic models or permitted exceptions, leaving the load-bearing factual illustration of the 'gap' partially untested against the cited events.
minor comments (2)
  1. [Abstract] The abstract and introduction should explicitly cross-reference the 2023 experiment's methodology and sample size when claiming 'roughly 50% divergence' and 'lower bias across nine social dimensions' to improve traceability.
  2. [Introduction] Notation for the 'political community deficit' is introduced without a formal definition or comparison table to related concepts such as 'democratic deficit' in EU studies; a brief definitional paragraph would aid clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these precise and constructive comments, which help clarify the normative and factual foundations of our analysis. We respond to each major comment below and indicate the revisions we will undertake.

read point-by-point responses

  1. Referee: [§4] §4 (Democratic Contestation): The central claim that the constitution's comprehensiveness 'forecloses democratic contestation' by resolving AI value questions rests on the unargued premise that such questions must remain open exclusively for public bodies; the manuscript does not demonstrate why corporate resolutions (even comprehensive ones) cannot serve as amendable baselines subject to regulatory override, market pressure, or later democratic input, as suggested by the skeptic's note on the 2023 experiment's potential as a contestable reference.

    Authors: We accept that §4 would be strengthened by an explicit defense of why corporate resolutions cannot serve even as provisional, amendable baselines. The manuscript's argument rests on the political community deficit: legitimacy requires initial authorization by a democratic body rather than post-hoc contestability through regulation or markets. We will revise §4 to elaborate this distinction, noting that the 2023 experiment's ~50% divergence demonstrates how corporate authorship systematically differs from public input, making corporate baselines non-neutral starting points. This revision will directly engage the skeptic's note while preserving the core claim that comprehensiveness risks foreclosing deliberation. revision: partial

  2. Referee: [§3] §3 (Military Gap): While the Palantir Maven integration during Iran strikes is cited as exposing the exclusion of military contexts, the analysis does not address whether the government-wide ban applied uniformly to all Anthropic models or permitted exceptions, leaving the load-bearing factual illustration of the 'gap' partially untested against the cited events.

    Authors: The referee correctly notes that the factual illustration could be more robust. The manuscript draws on contemporaneous public reporting of the government-wide ban and the documented continued use of Claude within Palantir's Maven platform during the Iran strikes. We will add a clarifying sentence in §3 stating that available records indicate the ban was framed as comprehensive with no publicly reported exceptions for military applications in this case. This will be incorporated as a minor factual elaboration without changing the substantive analysis of the constitution's exclusion. revision: partial

Circularity Check

0 steps flagged

No significant circularity in normative analysis

full rationale

The paper's derivation relies on external cited documents (Anthropic's 2026 constitution and 2023 participatory experiment) and standard democratic theory to support the 'political community deficit' claim. No self-definitional reductions, fitted predictions, or load-bearing self-citations appear; the argument draws on independent sources without author overlap or uniqueness theorems imported from prior work by the same authors. The central claims remain self-contained against external benchmarks rather than reducing to the paper's own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on domain assumptions about democratic legitimacy rather than empirical parameters or new entities with independent evidence.

axioms (1)
  • domain assumption Democratic legitimacy requires ongoing public deliberation on fundamental values and moral questions
    Invoked to support the claim that the constitution's resolution of issues forecloses contestation.
invented entities (1)
  • political community deficit no independent evidence
    purpose: To name the absence of a democratic body for setting AI governance principles
    New conceptual term introduced to frame the governance critique

pith-pipeline@v0.9.0 · 5493 in / 1156 out tokens · 43352 ms · 2026-05-13T18:55:46.814625+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

  1. [1]

    21, 2026), https://www.anthropic.com/constitution [https://perma.cc/Q7NC-NWFP] [hereinafter Anthropic Constitution]

    Claude’s Constitution , ANTHROPIC (Jan. 21, 2026), https://www.anthropic.com/constitution [https://perma.cc/Q7NC-NWFP] [hereinafter Anthropic Constitution]

    work page 2026

  2. [2]

    Claude’s Constitution , ANTHROPIC (May 9, 2023), https://www.anthropic.com/news/claudes-constitution [https://perma.cc/Y2KM-UDXX]

    work page 2023

  3. [3]

    G.A. Res. 217 (III) A, Universal Declaration of Human Rights (Dec. 10, 1948), https://www.un.org/en/about-us/universal-declaration-of-human-rights [https://perma.cc/ZU8R - 8SZR]

    work page 1948

  4. [4]

    Apple Media Services Terms and Conditions, APPLE, https://www.apple.com/legal/internet- services/itunes/us/terms.html [https://perma.cc/VP5Q-VWG2]

    work page

  5. [6]

    2 CALIFORNIA LAW REVIEW ONLINE [Vol

    Id. 2 CALIFORNIA LAW REVIEW ONLINE [Vol. 17:1 The constitution is not merely aspirational. In Anthropic’s earlier model, a shorter set of principles was used to train the AI through a process of automated feedback, rewarding outputs that aligned with the constitutional text and penalizing deviations. The 2026 constitution goes further. It is no longer a c...

    work page 2026

  6. [7]

    uniquely unsettling

    See HARV. BUS. SCH., A Fireside Chat with Sam Altman, OpenAI CEO at Harvard University (May 1, 2024), https://www.hbs.edu/about/video.aspx?v=1_mjf3unh9 [https://perma.cc/EV7M-3ZKE ] (stating that advertising in AI is “uniquely unsettling” and describing ads as “a last resort” business model for OpenAI); see also Natasha Lomas, Ads Might Be Coming to ChatG...

    work page 2024

  7. [8]

    wouldn’t necessarily be trained on the same c onstitution

    Our Approach to Advertising and Expanding Access , OPENAI (Jan. 16, 2026), https://openai.com/index/our-approach-to-advertising-and-expanding-access/ [https://perma.cc/6TRG- JSQK]. 2026] CORPORATIONS CONSTITUTING INTELLIGENCE 3 in part, because we learned long ago not to rely on the sustained benevolence of powerful private actors. But a possible future i...

    work page 2026

  8. [9]

    21, 2026), https://time.com/7354738/claude -constitution-ai-alignment/ [https://perma.cc/4JRZ-ETBB]

    Billy Perrigo, How Do You Teach an AI to Be Good? Anthropic Just Published Its Answer, TIME (Jan. 21, 2026), https://time.com/7354738/claude -constitution-ai-alignment/ [https://perma.cc/4JRZ-ETBB]

    work page arXiv 2026

  9. [11]

    any lawful purpose

    See Rebecca Bellan, Anthropic Sues Defense Department over Supply -Chain Risk Designation, TECHCRUNCH (Mar. 9, 2026), https://techcrunch.com/2026/03/09/anthropic -sues- defense-department-over-supply-chain-risk-designation/ [https://perma.cc/HJ26-Y7S7 ] (reporting that the Department of Defense wanted Anthropic to grant the agency unrestricted access to A...

    work page 2026

  10. [12]

    See id. (“Anthropic had two firm red lines: It didn’t want its technology to be used for mass surveillance of Americans and didn’t believe it was ready to power fully autonomous weapons with no humans making targeting and firing decisions.”). 4 CALIFORNIA LAW REVIEW ONLINE [Vol. 17:1 company ever to receive the label, a designation ordinarily reserved for...

    work page 2026

  11. [13]

    Anthropic is the only American company ever to be publicly named a supply chain risk, as the designation has traditionally been used against foreign adversaries

    See Jordan Novet, Anthropic Officially Told by DOD That It’s a Supply Chain Risk Even as Claude Used in Iran, CNBC (Mar. 5, 2026), https://www.cnbc.com/2026/03/05/anthropic-pentagon-ai- claude-iran.html [https://perma.cc/D7AC-SBSS] (“Anthropic is the only American company ever to be publicly named a supply chain risk, as the designation has traditionally ...

    work page 2026

  12. [14]

    unprecedented and unlawful

    See Bellan, supra note 15 (reporting that Anthropic filed two complaints in California and Washington, D.C., calling the DOD’s actions “unprecedented and unlawful”)

    work page

  13. [15]

    Anthropic’s Claude has become a crucial component of Palantir’s Maven intelligence analysis program

    See Dan De Luce, Gordan Lubold, Kevin Collier & Jared Perlo, U.S. Military Is Using AI to Help Plan Iran Air Attacks, Sources Say, as Lawmakers Call for Oversight, NBC NEWS (Mar. 12, 2026), https://www.nbcnews.com/tech/tech -news/us-military-using-ai-help-plan-iran-air-attacks- sources-say-lawmakers-rcna262150 [https://perma.cc/E2XT-RJZP] (“Anthropic’s Cl...

    work page 2026

  14. [16]

    the U.S. used Anthropic ’s Claude AI model . . . for the attack on Iran—and is still using it

    See James LaPorta & Camilla Schick, Anthropic’s Claude AI Being Used in Iran War by U.S. Military, Sources Say, CBS NEWS (Mar. 3, 2026), https://www.cbsnews.com/news/anthropic- claude-ai-iran-war-u-s/ [https://perma.cc/7DZ2-865F ] (confirming that “the U.S. used Anthropic ’s Claude AI model . . . for the attack on Iran—and is still using it” despite “a go...

    work page 2026

  15. [17]

    Anthropic Constitution, supra note 1

    work page

  16. [18]

    Gilad Abiri, Public Constitutional AI, 59 GA. L. REV. 601 (2025)

    work page 2025

  17. [19]

    17, 2023), https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language- model-with-public-input [https://perma.cc/4WZ6-ZJUF]

    Collective Constitutional AI: Aligning a Language Model with Public Input, ANTHROPIC (Oct. 17, 2023), https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language- model-with-public-input [https://perma.cc/4WZ6-ZJUF]

    work page 2023

  18. [20]

    lower bias across nine social dimensions

    Id. 6 CALIFORNIA LAW REVIEW ONLINE [Vol. 17:1 principles showed “lower bias across nine social dimensions” while “maintaining equivalent performance” on language and reasoning tasks.28 Anthropic proved that participatory constitution -making for AI is technically feasible. They demonstrated that it produces measurably different, and in some respects bette...

    work page

  19. [21]

    Saffron Huang et al., Collective Constitutional AI: Aligning a Language Model with Public Input, in PROCEEDINGS OF THE 2024 ACM CONFERENCE ON FAIRNESS, ACCOUNTABILITY, AND TRANSPARENCY (FAccT ‘24) 1395 (2024), https://doi.org/10.48550/arXiv.2406.07814 [https://perma.cc/R4UB-S68X]

    work page doi:10.48550/arxiv.2406.07814 2024

  20. [22]

    Abiri, supra note 22

    work page