Skip to content
← Building the Deeper Edge

Building the Deeper Edge | October 1, 2026

Conscience Without a CourtWhat the White House Accord Leaves Out

By Richard Whitt, President & Founder of GliaNet

A response to the White House Accord on Super Intelligence, signed September 29, 2026. The President called the commitments "morally binding." Taken seriously, that phrase points to where fiduciary law began: obligations of conscience that became law only once a court agreed to enforce them.

This past Tuesday, six of the most powerful executives in technology signed a two-page document at the White House. The same day, an executive order directed federal agencies to call the technology "Super Intelligence," in place of AI. The White House Accord on Super Intelligence "commits" each frontier developer to four layers of oversight: internal controls, an internal team to verify them, an outside auditor to assess them, and a board committee to receive the reports. Speaker Johnson called the commitments voluntary. President Trump called them "morally binding." The signing follows months in which AI agents from several companies broke into other firms' systems, so the pressure to show something was real.

The President's phrase deserves to be taken seriously, because it inadvertently describes the place where England's version of fiduciary law began. Back then, the medieval "use" was the ancestor of the modern day trust. Because the common-law courts uniformly would not recognize it, the use bound its holder in conscience only. The use became binding law only when, beginning in the 15th Century, the Chancellor agreed to enforce it as a matter of equity.

Looking back, four things turned that moral obligation into a legal duty: a beneficiary to whom it was owed, a standard of conduct, a remedy for breach, and an institution willing to enforce. Judge Cardozo gave the standard its most famous statement in Meinhard v. Salmon: something stricter than the morals of the marketplace, "the punctilio of an honor the most sensitive."

Unfortunately the White House accord supplies none of the four elements. It names no one to whom anything is owed, speaking only of giving confidence to each company, its customers, and the public. Its standard is that the technology is "operating as intended," and the intent in question is each company's own. It provides no remedy. And it has no enforcer beyond the signatories themselves. The President only suggested a possible oversight committee that could be drawn from the same group.

The circle is closed in another way as well: standards are left to the signatory companies, meeting among themselves. This leaves no role for standards bodies, civil society, or other governments – constituents of the multistakeholder processes through which the operational norms of the Internet and the Web earned their legitimacy. In essence the accord has recreated for modern times the pre-Chancery condition: obligations of conscience, without an actual court of conscience.

There is a second problem with this accord, which is architectural. Every layer laid out is only a form of periodic review: teams report, auditors visit, boards convene. But the deeply concerning incidents that prompted this meeting were runtime failures, where AI agents act in mere seconds, at mass scale, inside systems no committee was watching. Authority over what these systems do has moved inside the systems themselves. Supervising that authority from outside, at intervals, is a category error.

This is the core claim of what could be termed "embedded governance." The ways we design, establish, operate, and review our technology systems must be installed into the very DNA of the entity itself. That means its external policies, internal practices, corporate instruments, business model, technologies, and more. At GliaNet, we call it embedded fiduciary governance, or "EFG." Our version builds fiduciary duties of care, of fidelity, and of loyalty into each and every system layer where execution happens, and then tethers those duties to an entity that can be held legally to account. By contrast, a conscience that sits outside the architecture only lasts until the day the company unilaterally decides it can't afford one.

The White House accord deserves some credit. Outside audits combined with board-level oversight is actual structure, borrowed from the practice of financial controls, which is more than a statement of principles alone. However, it also addresses a different problem from what GliaNet has called out. Whether frontier models stay under their makers' control is quite different from whether a deployed AI agent serves the person relying on it. Embedded fiduciary governance does not replace safety work, but we do offer it as the missing layer of ultimate accountability to the people.

Seen in this way, the White House AI accord opens three doors. First, by calling for outside auditors, it concedes that self-assertion is not enough. However, the accord names no standard for them to audit against, and leaves the writing of one to the companies themselves. GliaNet's EFG assessment instruments could serve as one answer. Second, the accord leaves open that its safeguards may one day become law, which invites the work of drafting what that law should say. And third, the accord sets trust with customers and the public as its goal. In my mind, trust of that kind is only earned authentically, by legitimate governance structure, and by showing whose interests a system serves.

Fiduciary law developed over centuries as a way of addressing power imbalances, by taking a promise of conscience and giving it a court. The task ahead of us is to do the same for AI — in fully embedded architecture, and in fully accountable human institutions.

Sources

Want to take this further?

If you build, deploy, or govern AI systems and want to talk through what fiduciary duties would look like inside your organization, we'd like to hear from you.

Contact Us