What the SEC Presto Order Means for Human-in-the-Loop (HITL) AI Compliance

The SEC’s January 2025 enforcement action against Presto Automation exposes the compliance gap between marketing "autonomous AI" and the reality of relying on hidden human labor. By mapping Presto’s disclosure failures to Lozen Advisory’s Name Standard℠ and Disclosure-Independent Governance℠ methodology, this analysis shows why corporate boards need an active, empowered, documented human accountability layer and clear committee ownership for AI claims.

AI Accountability: Beyond Human-in-the-Loop

This article uses the SEC’s January 2025 enforcement action against Presto Automation as a case study in AI accountability failure, and maps each violation to Lozen Advisory’s Name Standard℠ within the firm’s Disclosure-Independent Governance℠ methodology. The Presto action is not only an AI-washing case. It is a disclosure-control case about ownership, human intervention, and whether a company can describe an AI product as autonomous, specifically, as having eliminated human order taking, when a human-in-the-loop (HITL) workforce remains embedded in the operating model.

On January 14, 2025, the SEC announced settled charges against Presto Automation, Inc. for making materially false and misleading statements about Presto Voice, its drive-thru AI ordering product. The enforcement signal is direct: public-company AI claims now require disclosure controls that can verify the difference between what the company says AI does, and which human roles, vendors, or operational functions are actually involved in making the AI work. For corporate boards, that verification problem also creates a committee ownership problem: someone must know which board body owns the AI claim, the vendor dependency, human review layer, and the escalation path.

Presto Claimed the Technology and Erased the Humans

The SEC order identifies two false claims and a third structural failure.

• False claim one: From November 2021 to September 2022, Presto described Presto Voice as “our technology” and “Presto’s technology” while failing to disclose that every commercially deployed unit was powered by technology owned and operated by Hi Auto, a third-party supplier.
• False claim two: After deploying its own proprietary technology in September 2022, Presto claimed Presto Voice “eliminates human order taking.” The original version required human agent intervention on every order. A more advanced version, piloted June 2023, still required human intervention approximately 70 percent of the time. The agents were located primarily in the Philippines and India.
• Third violation: From September 2022 through December 2024, Presto had no disclosure controls or procedures. No one at the company was formally responsible for ensuring that its Commission filings were accurate.

The performance data Presto reported compounded the misrepresentation. The “automated order completion” and “non-intervention” rates of 95 to 99 percent measured orders completed without restaurant staff involvement, not without any human involvement. The offshore agents were excluded from the count entirely. When Presto later disclosed a system-wide non-intervention rate of 85 percent, that figure still used the same definition: no restaurant staff involvement. The actual rate of orders completed without any human involvement was lower, and at the original-version locations, where human agent intervention was required on every order, it was effectively zero.

Presto’s own executives knew this from the beginning. On January 20, 2022, months before Presto began making autonomous-product claims, a Presto executive messaged another explaining that “with HITL [humans in the loop], accuracy is not a major concern” and “can even get to 95% or more with humans.” The following day, the same executive wrote that Presto would “only be transitioning to our own Presto solution when we’ve nailed the HITL [humans in the loop] and have comparable accuracy / latency.” Executives knew the reported accuracy figures were HITL-dependent. In October 2022 a senior executive warned internally not to use “automation rate with customers because it infers no supervision which isn’t true.” In January 2023 another raised the alarm that the company was “telling investors Presto AI is running 95%+ accuracy without disclosing AI is doing NONE of the work and all orders are processed by humans.” No corrective disclosure followed until after the Commission’s investigation began.

Full findings: SEC Order, File No. 3-22413 (Jan. 14, 2025)

Why This Is a Name Standard℠ Failure Case

The Name Standard℠ asks a single governance question: when AI-assisted work moves through a company’s name, a product’s name, or a human’s professional authority, has the accountable human contribution been preserved, verified, and disclosed where material?

Human-in-the-loop and human-on-the-loop describe placement — where a person sits in the workflow relative to AI output. They do not answer the attribution question. Before a human’s name, judgment, or institutional sign-off is attached to AI-assisted output, does that person have the capacity to verify it, the information required to evaluate it, the authority to reject it, and a formal escalation path if something is wrong?

Placement is not accountability. The Name Standard℠ evaluates accountability.

What follows is Lozen Advisory’s diagnostic reading of the Presto order — not additional SEC findings — applied to show where a structured human accountability framework would have intervened.

• Pillar 1 — Time Allocation. Does the responsible individual have the explicit operational hours required to verify complex synthetic outputs? In Presto’s case, that means verifying AI-product claims, operational metrics, and human-intervention rates against actual workflow before filing. No one was formally assigned to do that. The claim that Presto Voice “eliminates human order taking” appeared in five registration statements over seven months without a review cycle that caught the gap.
• Pillar 2 — Review Capacity and Tools. Does the reviewer possess the necessary diagnostic source data to actively validate the AI’s assertions? Presto’s reported non-intervention rates of 95 to 99 percent were never reconciled against actual operational data before filing. Executives raised the terminology problem internally — but no disclosure-control process required the company to resolve the gap between operational reality, internal warnings, and public claims before those claims reached investors.
• Pillar 3 — Information Access. Is there a clear, uninhibited line of sight into content provenance and vendor-utilized foundation models? In this case, the equivalent question is whether anyone with disclosure responsibility had documented visibility into the supplier technology powering the product. Presto’s reliance on Hi Auto — which owned and operated the AI technology powering every deployed unit — was never adequately disclosed. Did anyone with disclosure responsibility have clear, documented visibility into that dependency or its materiality to investors?
• Pillar 4 — Documentation Infrastructure. Is the human validation process tangibly logged to provide clear internal compliance evidence during regulatory review? This is the third violation in the SEC order in explicit form. Presto had no established process for drafting, reviewing, or approving Commission filings, and no one was formally responsible for ensuring their accuracy. The order found an absence of disclosure controls and procedures — meaning no formal validation chain existed to produce or preserve a compliance record.
• Pillar 5 — Formal Right of Refusal. Does the individual possess the institutional standing and explicit veto power to halt, escalate, or document non-compliant AI assets without fear of internal reprisal? Presto executives flagged the misleading terminology in October 2022 and again in January 2023. Those concerns produced internal discussion and no corrective action. Where was the formal escalation path for the people who knew the claims were false to stop the filings.

The Name Standard℠ does not ask whether AI is impressive. It asks whether the human accountability layer is active, documented, and empowered or ceremonial. This is a Disclosure-Independent Governance℠ problem because the company’s SEC filings claimed autonomous AI performance while omitting the human intervention and vendor dependency required to make the product function.

The Governance Signal for Boards

The Presto order is board-relevant. It is an issuer disclosure enforcement action — which means disclosure controls, AI claim accuracy, investor materiality, and human accountability are the questions boards and audit committees are responsible for. The five pillars of the Name Standard℠ are the diagnostic — examined in full in Lozen Advisory’s Disclosure-Independent Governance℠ category framework, which maps these accountability failures to the board evidence problem AI adoption is creating across enterprises. The Committee Ownership Map extends that question by asking which board body owns AI-assisted decision accountability, vendor reliance, escalation duties, and evidence of human sign-off. The Presto order illustrates what can happen when none of them are in place.

For boards, audit committees, and disclosure teams, the question is not whether your company uses AI. It is whether the human-in-the-loop (HITL) layer behind your AI claims is active, documented, and empowered to act — or whether it exists only on paper.

The 5Ws of Decision Integrity℠ — Lozen Advisory’s board-level diagnostic — translates that question into five evidence standards boards can apply to any significant AI-assisted decision or AI product claim:

• What data, models, or AI outputs did management rely on, and is that reliance documented?
• Who held named accountability for this decision, and did that person have the authority and information required to sign off?
• Which performance indicators were tracked, and what would trigger board escalation if the system degraded or produced misleading output?
• What was missed — what risks, gaps, or AI errors were not detected, and who is responsible for that gap?
• Witness: What documented record exists that would withstand regulatory scrutiny, litigation, or board inquiry?

At Presto, none of these questions had a documented answer.

Presto’s executives knew the claims were false. The problem was not awareness. It was the absence of any formal structure — time, tools, access, documentation, or veto authority — to translate that awareness into corrective disclosure. That absence is the third violation in the order: a violation of Exchange Act Rule 13a-15(a), which requires issuers to maintain disclosure controls and procedures. The order also found violations of Section 17(a)(2) of the Securities Act, Section 13(a) of the Exchange Act, and Rule 13a-11 for the misleading filings themselves.

The SEC Wrote Humans Back Into AI Disclosure

The Presto case is powerful because the enforcement theory returns humans to the center of the AI story.

The company claimed AI eliminated the need for human order taking.

The SEC found that the original version of Presto’s proprietary AI required human intervention on every order. Even the more advanced version, piloted in June 2023, still required human agent involvement 70 percent of the time.

That is the Name Standard℠ argument in enforcement form.

When AI carries a company’s name, humans cannot be treated as invisible if that human layer is still necessary to make the system function. The offshore agent in the Philippines or India entering the drive-thru order is not an implementation detail. If people are still reviewing, correcting, completing, approving, escalating, or taking the order — if the system cannot function without them — the company has not eliminated human intervention. It has only changed where the human sits and whether investors can see them.

Presto wrote the humans out.

The SEC wrote them back in.

Board AI Algorithmic Accountability

Organizations facing AI-related disclosure scrutiny, evidence failures, or reputational exposure can use Lozen Advisory’s Board AI Algorithmic Accountability service to reconstruct what the record can prove and where accountability remained unresolved.