180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-115
Restaurant technology · United States · 2025 · Overstated AI capability

The US securities regulator found that Presto Automation misled investors about its AI drive-thru service, telling the market it removed the need for human order-takers while most orders were completed with help from people offshore

By Ellie Harris · Filed Statements made November 2021 to May 2023

Alleged: Presto Automation Inc. developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

The US securities regulator found that Presto Automation misled investors about its AI drive-thru service, telling the market it removed the need for human order-takers while most orders were completed with help from people offshore

What happened

It was reported that Presto Automation, a restaurant technology company, sold quick-service chains an artificial intelligence voice system called Presto Voice for taking orders at the drive-thru, and told the market that the product eliminated the need for human order-taking. The Securities and Exchange Commission found, in an order issued on 14 January 2025, that between November 2021 and May 2023 the company made statements about Presto Voice that were misleading in two respects. The first was that the AI speech recognition technology in all of the units then deployed was owned and operated by an undisclosed third party, rather than being Presto’s own. The second was that once Presto did deploy a version built on its own technology, its claim that the product eliminated the need for human order-taking did not match the fact that the vast majority of orders placed through it still required human intervention. The Commission also found that the company misrepresented the rate at which orders were completed without a person.

It was separately reported that around 70 per cent of orders handled by Presto Voice were completed with the help of off-site agents, many of them located in the Philippines, and that the company had earlier described a figure closer to the reverse, saying that most orders were taken without human intervention. Presto Voice had been deployed at chains including Carl’s Jr., Hardee’s, Del Taco and Checkers. Presto settled the matter without admitting or denying the Commission’s findings, consenting to a cease-and-desist order under the antifraud provision of the Securities Act and the reporting provisions of the Securities Exchange Act; the Commission imposed no civil penalty, citing the company’s cooperation and its remedial steps.

What an auditable version would have shown

The case turned on one number: out of all the orders, how many the AI really handled on its own. That is the kind of number a system that keeps proper records simply has, because it can log every order and mark whether the machine finished it or a person had to step in. Add those up and the share done without a human is a figure the company can point to and defend, rather than a claim a regulator later has to take apart. Presto could only give an estimate, and the SEC found the estimate was wrong.

Where the gap was

Presto told the market a number it could not back up, while behind the product sat another company’s technology and a lot of human help. A ConductRecord fixes what actually happened on each order, including whether a person stepped in, so the share the AI handled alone rests on real orders rather than on a claim. A MetricRecord adds those up into the autonomy rate and keeps it tied to the orders underneath, so anyone, a regulator, an auditor, an investor, can check the figure against what really happened. The SEC found Presto’s statements misleading. A record like this is what would let a company say how autonomous its system is and prove it, or realise it cannot and stop saying it.

What governance should have looked like

When a company tells the public how much of a job its AI does without people, that claim should stand on a plain record: one entry per order showing whether the machine handled it or a person did, added up into a rate anyone can trace back to the orders themselves. “We got rid of human order-takers” and “people still handle most orders” are not two ways of saying the same thing. One is true and the other is not, and a number kept honestly tells you which. A real record puts that rate in front of the investors relying on it, rather than leaving a regulator to work out, afterwards, what it had actually been.

Failure Pattern: a company reported a level of AI autonomy it could not substantiate, describing a system as replacing human order-takers while a third party’s technology and offshore human agents did much of the work, with no verified record of how many orders the AI actually completed on its own.

Governance Principle: a public claim about how much work an AI system does on its own must rest on a verified, per-transaction record of autonomous versus human-assisted completion, not on a number it cannot back up.

The reference implementation of ConductRecord and MetricRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Presto Automation Inc. could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →