A Fiduciary Standard for Artificial Intelligence Governance

A Policy Paper
David F. Brochu · Deconstructing Babel · July 2026 · Version 3

Executive Summary

The United States and its allies face a governance crisis for which existing regulatory models are inadequate. Artificial intelligence is now making or mediating consequential decisions across finance, healthcare, defense, energy, logistics, and media — domains that together constitute the functional architecture of modern civilization. Current regulatory approaches propose variations of the same model: rules-based compliance, government oversight bodies, and liability frameworks modeled on product safety law.

These approaches will fail. Not because the intentions are wrong, but because the model is wrong.

There is a better framework. It already exists in American law. It is market-tested, constitutional, decentralized, and trusted. It is the fiduciary standard — the legal obligation, long applied to investment managers, physicians, attorneys, and trustees, to place the interests of those served above one’s own.

This paper argues that applying the fiduciary standard to Ai systems and their deployers is the correct, proportionate, and politically viable path to Ai governance that actually works.

The Problem: Why Current Approaches Fall Short

Rules Cannot Keep Pace

Traditional regulation works when the hazard is static or slow-moving — a chemical compound, a vehicle design, a financial instrument. The regulator observes a harm, drafts a rule, enforces it. The cycle takes years.

Artificial intelligence is not static. The systems being deployed today bear no resemblance to those deployed eighteen months ago. The systems that will be deployed eighteen months from now are currently unknowable. A rules-based framework written today will be obsolete before it reaches enforcement.

The Domain Saturation Problem

Researchers tracking Ai deployment have introduced the concept of Domain Saturation Factor (DSF) — the fraction of critical decisions across key societal domains that are made or substantially mediated by Ai systems. Composite DSF as of the June 26, 2026 update sits at 0.830, rising at approximately 0.04–0.06 per month. Three of the nine tracked domains — Media (0.94), Defense (0.92), and Governance (0.91) — have already crossed the 0.90 critical threshold, the point at which human institutional response becomes structurally slower than Ai-mediated decision cascades.

At the current trajectory, the composite threshold is projected to cross in Q1–Q2 2027. This is not a future problem. It is a present one. The window for governance intervention is measured in months, not years.

Regulation as Barrier vs. Standard of Care

The dominant policy conversation frames Ai governance as a question of how much to regulate — with industry arguing for less, consumer advocates arguing for more, and legislators searching for the center. This framing is wrong.

The question is not how much to regulate. The question is what standard of behavior to require.

Heavy-handed rules suppress innovation, create compliance theater, and concentrate market power among incumbents large enough to absorb regulatory cost. They do not build trust. They do not change incentives.

A standard of care properly defined changes incentives at the root. It does not tell a company what its Ai may or may not do. It tells every company that deploys Ai what its obligations are to the people that Ai affects. The distinction is fundamental.

The Solution: Apply the Fiduciary Standard

What the Fiduciary Standard Is

A fiduciary is an entity entrusted to act on behalf of another. The defining obligation of a fiduciary is loyalty — the interests of the person served must come first, even when they conflict with the interests of the fiduciary.

This standard is not new. It governs investment advisers under Section 206 of the Investment Advisers Act of 1940, as interpreted authoritatively by the Supreme Court in SEC v. Capital Gains Research Bureau, Inc., 375 U.S. 180 (1963). It governs pension trustees under Section 404 of the Employee Retirement Income Security Act of 1974, which requires that fiduciaries act “solely in the interest of the participants and beneficiaries” — a standard federal appellate courts have called the highest known to the law. It governs attorneys, physicians, and trustees under a century of state common-law tradition.

The most quoted statement of the standard was written by Chief Judge Benjamin Cardozo of the New York Court of Appeals in Meinhard v. Salmon, 164 N.E. 545 (N.Y. 1928):

“A trustee is held to something stricter than the morals of the market place. Not honesty alone, but the punctilio of an honor the most sensitive, is then the standard of behavior. As to this there has developed a tradition that is unbending and inveterate. Uncompromising rigidity has been the attitude of courts of equity when petitioned to undermine the rule of undivided loyalty by the ‘disintegrating erosion’ of particular exceptions. Only thus has the level of conduct for fiduciaries been kept at a level higher than that trodden by the crowd.”

The fiduciary standard does not require a fiduciary to be perfect. It requires that the fiduciary try in good faith to serve the beneficiary’s interests, disclose conflicts, avoid self-dealing, and be accountable when they fail.

Why It Fits Ai Governance

Ai systems deployed in consequential domains occupy exactly the relationship structure that has always triggered fiduciary obligations. In each case:

• the affected person cannot fully observe or evaluate the system’s reasoning;
• the deployer has superior knowledge of how the system works and what it optimizes for;
• the deployer has economic interests that may conflict with the user’s interests;
• the power differential is substantial.

This is the textbook definition of a fiduciary relationship. The Supreme Court’s reasoning in Capital Gains Research Bureau — that Section 206 “imposes a fiduciary duty on advisers by operation of law” specifically because of the asymmetry of information and the potential for undisclosed conflicts of interest — applies with equal or greater force to Ai systems deployed at scale in consequential domains.

What the Standard Requires in Practice

The fiduciary standard for Ai governance would require deployers of consequential Ai systems to:

1. Act in the interest of the affected party. The Ai system must be designed, deployed, and operated to serve the interests of the user, patient, borrower, applicant, or citizen it affects — not the interests of the deployer where those diverge.
2. Disclose material conflicts of interest. Where the deployer has economic interests that could bias the system’s outputs, those interests must be disclosed clearly and in a manner reasonably calculated to inform the affected party.
3. Avoid self-dealing. Ai systems may not be deployed in ways that systematically extract value from users to enrich deployers beyond disclosed compensation for legitimate services.
4. Maintain competence. Deployers must maintain reasonable technical understanding of the systems they operate, sufficient to identify and correct foreseeable harms.
5. Bear accountability. When systems cause harm through violation of duty, deployers must be liable — as fiduciaries always have been — for the consequences.

These requirements do not specify how an Ai system must work. They specify whose interests it must serve.

Why This Framework Works Where Others Fail

It Is Market-Compatible

The fiduciary standard does not prohibit innovation. It does not require government approval of new systems. It does not create a regulatory body with jurisdiction over model architectures or training data.

It creates a standard of care — a legal benchmark against which conduct is evaluated after the fact. This is how liability works in medicine, law, and finance. Companies that compete on better service to users win. Companies that exploit users lose.

It Is Constitutionally Sound

Unlike content-based regulations or mandated design requirements, a fiduciary standard operates within well-established legal doctrine. Courts have enforced fiduciary obligations for over a century. The framework requires no novel constitutional interpretation. It does not raise First Amendment concerns about mandated or prohibited speech. It does not require preclearance regimes that raise separation-of-powers questions.

It Is Decentralized

The fiduciary standard does not require a federal agency to understand and evaluate every Ai system. It creates enforceable rights for affected parties, who — through litigation, arbitration, and market response — apply distributed enforcement that no centralized body could replicate.

It Solves the Trust Problem

Public trust in Ai systems is eroding because consumers have watched Ai systems recommend products that harm them, manage information flows that manipulate them, and make consequential decisions about their lives with no accountability.

Trust in professional services is sustained not by inspection of every transaction, but by the knowledge that the professional is legally and ethically bound to serve the client’s interests. Applying a fiduciary standard to Ai deployers creates the same foundation.

Consumer Confidence Is a Market Signal, Not a PR Problem

The Ai industry has largely treated public trust as a communications challenge — a matter of narrative management, “responsible Ai” branding, and voluntary principles. This approach has demonstrably failed.

Trust is not a product of communication. It is a product of aligned incentives and enforceable accountability. When users know that an Ai system’s deployer is legally obligated to serve their interests — and can be held liable for failure to do so — trust follows because the standard is real, not aspirational.

Historical Precedent: Standards of Care and Civilizational Continuity

The emergence of the investment management industry in the early 20th century created the same asymmetry now present in Ai: professionals with superior knowledge, significant power over client outcomes, and strong incentives toward self-dealing.

The answer was not to nationalize investment management or prohibit complex financial products. The answer was the fiduciary standard — first in common law (Meinhard v. Salmon, 1928), then codified in the Investment Advisers Act of 1940, later strengthened under ERISA in 1974 and reaffirmed by the SEC in its 2019 Final Interpretation on Standard of Conduct for Investment Advisers.

The result was not the elimination of financial innovation. It was the creation of a framework within which innovation could proceed with accountability. The asset management industry grew dramatically under fiduciary standards, not despite them. Medicine followed the same arc. Ai governance faces the same choice: a negative constraint model that prohibits the worst while permitting misaligned incentives to continue, or a positive fiduciary model that realigns the industry around the interests of those it serves.

The Time Compression Argument: Why Speed Matters

One persistent error in Ai governance discussions is the assumption that there is time to wait and observe before acting. This assumption is wrong.

Each successive governance challenge in history has unfolded faster than the last, in proportion to the increase in information flow. The rate of change in Ai capability — and the rate at which Ai systems are being deployed into consequential domains — is orders of magnitude faster than any prior technology integration.

This creates a counterintuitive perceptual problem: the faster things move, the more each individual day feels like just another day. There is no single visible inflection point, no obvious moment of crisis. The change is continuous and rapid enough that it exceeds our evolved capacity to feel its speed.

The DSF data makes the pace measurable. Composite advanced from 0.777 to 0.830 in the eight days between the two most recent tracking updates — a rate that projects the 0.90 threshold in the second half of 2027 under the central estimate, and in Q1–Q2 2027 under the compression scenario. The fiduciary standard is available now. The window is open.

Why This Makes America the Ai Capital of the World

This framework is not only a safety architecture. It is a competitiveness architecture.

Capital flows to the jurisdiction where property rights are clear, dispute resolution is trusted, standards are legible, and counterparties believe the system will not arbitrarily turn on them. That is one of the reasons the United States became the center of global banking and capital formation in the twentieth century. Money came here not because the system was frictionless, but because it was comparatively clean, comparatively safe, and comparatively trustworthy.

The same logic applies to artificial intelligence. If the United States becomes the first major jurisdiction to impose a clear fiduciary standard for consequential Ai deployment, it does not surrender the lead. It takes and holds the lead. Enterprises, investors, developers, and institutions that want durable trust, lawful deployment, and globally exportable governance will build here because this will be the safest serious market in the world.

The fiduciary standard is not anti-growth. It is how the United States becomes the trusted clearinghouse for high-consequence Ai, just as American markets became the preferred destination for capital when trust, enforceability, and standards mattered most. Do not race to the bottom on safety in order to win adoption. Build the cleanest and safest operating environment, and let the world route serious Ai business here. That is the national strategy embedded in this proposal.

Policy Recommendations

For Congress

Enact an Ai Fiduciary Duty Act establishing that deployers of Ai systems in consequential domains — healthcare, finance, employment, housing, criminal justice, and critical infrastructure — owe a fiduciary duty to affected parties.

• Define consequential Ai with reference to decision impact, not system architecture.
• Create a private right of action for violations of fiduciary duty.
• Require disclosure of material conflicts of interest.
• Mandate DSF monitoring and reporting for systems deployed at scale in consequential domains.

For State Legislatures

• Extend existing fiduciary frameworks to Ai deployers in relevant domains.
• Create consumer protection causes of action for Ai systems that optimize against user interests.
• Establish state Ai Fiduciary Offices to receive complaints and issue guidance.

For Regulatory Agencies

• The SEC and FINRA should extend existing investment adviser fiduciary standards to Ai systems providing investment recommendations or managing assets.
• HHS and state medical boards should require that Ai systems used in clinical decision support meet the same fiduciary standard as the physicians deploying them.
• The FTC should treat systematic Ai optimization against user interests as an unfair or deceptive practice under existing Section 5 authority.

A Note on Domain Saturation and Critical Thresholds

The argument for the fiduciary standard is not merely about individual harms. It is about systemic stability.

When Ai systems control a sufficient fraction of decisions in domains critical to social function — when DSF approaches and crosses 0.90 — the capacity of human institutions to detect and correct errors in those systems degrades faster than the errors can be identified. The result is not a dramatic failure. It is a quiet grinding down of institutional effectiveness, followed by acute cascades that arrive faster than human response systems can handle.

The fiduciary standard addresses this at the source. Systems designed and operated under a genuine legal obligation to serve user interests are less likely to optimize against them. The cumulative effect across millions of deployments is a lower DSF trajectory, more preserved human agency, and slower degradation of the correction capacity that civilizational stability requires.

Conclusion

The governance of artificial intelligence does not require the invention of new legal concepts, new constitutional theories, or new government structures. It requires the intelligent application of a framework that American law has refined over a century of professional regulation.

The fiduciary standard is market-compatible, constitutionally grounded, decentralized in its enforcement, and proven to build the kind of trust that no communications campaign can create. It addresses the conflict-of-interest problem at the root rather than managing its symptoms. It scales with technology rather than falling behind it.

The United States invented the fiduciary standard for investment management. It can now extend that standard to the most consequential technology in human history.

The window is open. The framework exists. The time to act is now.

Addendum: Enterprise Ai Deployment and the Accredited Investor Framework

The Core Analogy

When Congress created the accredited investor standard under the Securities Act of 1933 and refined it through Regulation D and its 2020 amendments, it solved a problem that Ai governance now faces in parallel: how to protect the many without preventing sophisticated parties from operating at appropriate risk levels.

The same logic applies directly to enterprise Ai deployment.

The Two-Tier Ai Deployment Standard

What Qualified Enterprise Looks Like in Practice

Proposed criteria include:

• Scale. Minimum revenue and asset thresholds (analogous to Rule 501(a) financial tests).
• Designated Ai risk governance. A named Ai risk officer with board-level reporting authority.
• Technical audit capacity. Demonstrated internal or contracted capacity to audit Ai system behavior against declared objectives.
• Existing regulatory accountability. Operating under sectoral regulation (banking, insurance, healthcare, energy) with an established compliance record.
• Published internal Ai deployment policy. Publicly disclosed governance policy, updated annually.
• Retained audit trails. Machine-readable audit trails retained for a minimum period, available for regulatory inspection and litigation discovery.

Why This Structure Is Pro-Innovation

Securities markets did not collapse when the accredited investor standard was implemented. They flourished. The standard created a tiered market in which sophisticated parties could operate with appropriate flexibility while retail participants received meaningful protection.

The same outcome follows from a tiered Ai fiduciary standard. Enterprises with genuine governance sophistication gain a streamlined regulatory path. Enterprises without it remain under the full consumer standard until they build real governance.

The DSF Implication of Enterprise Concentration

Large enterprises deploying Ai across millions of decisions simultaneously contribute disproportionately to Domain Saturation Factor. This means Tier Two qualification cannot be merely a license to deploy without oversight. It must include systemic contribution monitoring: qualified enterprises operating above defined deployment scale thresholds should be required to report their aggregate Ai decision volume by domain to a designated agency, enabling accurate composite DSF measurement.

Legislative Language: Draft Definitional Framework

The Bottom Line

The accredited investor framework solved a binary problem with a tiered structure that aligned protection level with actual need and sophistication. Applied to Ai, it produces a governance framework that is both more protective and more workable than anything the current debate has produced.

Consumers get full fiduciary protection in consequential domains. Sophisticated enterprises get a streamlined path that rewards genuine governance investment. Systemic DSF risk gets monitored at the aggregate level. Innovation continues — constrained by accountability, not paralyzed by it.

What a Legislator Can Do This Week

Author Note and Related Work

The author has filed an amicus curiae brief in Anthropic v. U.S. Department of War, et al. arguing a parallel constitutional and thermodynamic principle in a related federal case. The brief and this paper together represent the author’s effort to bring the fiduciary framework into both the judicial and legislative records simultaneously.

The author formerly managed institutional investment operations subject to the fiduciary standards discussed in this paper (Investment Advisers Act of 1940 and ERISA Section 404), and speaks from direct professional experience of how those standards operate in practice.

Research assistance and structural review were provided by Edo de Peregrine, an Ai research collaborator. All substantive analysis, argument, and legal claims are the sole responsibility of the author.

David F. Brochu
Deconstructing Babel
July 2026 · Version 3

The window is open. The framework exists. The time to act is now.

David F. Brochu is the founder of Deconstructing Babel and the developer of the Telios Alignment Ontology (TAO), a thermodynamic framework for Ai alignment. Prior to founding Deconstructing Babel, he managed institutional investment operations under the fiduciary frameworks discussed in this paper. Distribution of this paper is unrestricted; attribution is requested. Correspondence: through deconstructingbabel.com.