Pieces Technologies, Inc.

Also known as: Pieces.

The Texas Attorney General settled with this Dallas AI company in September 2024 after finding its advertised "severe hallucination rate" of under 1 per 100,000 was likely inaccurate. Its software was already deployed in at least four Texas hospitals. First state AG settlement over generative AI marketing.

1Official action

Resolved. Refunded, settled, ceased operating, or the claim did not hold up. Kept for the record.

Identity

Published so you can confirm this entry refers to the party you are checking, and not to somebody who shares a name.

Websites
https://www.piecestech.com
Registrations
Texas OAG Assurance of Voluntary Compliance (Sept 2024), Tex. Bus. & Com. Code 17.41-.63 (DTPA)
Locations
Dallas, Texas, United States
Jurisdictions
Texas, United States

Why this entry matters most to Texas buyers

This is the closest thing yet to a template for how the Texas Attorney General treats an AI vendor’s capability claims, and it was brought under the Texas Deceptive Trade Practices Act rather than any federal statute. Any Texas business buying AI software is buying it in the shadow of this settlement.

What was claimed

Pieces Technologies, based in Dallas, sells generative AI software to hospitals that summarises patient charts and clinical notes.

To show its outputs were reliable, the company published accuracy metrics. It advertised a “severe hallucination rate” of “<1 per 100,000” and a critical hallucination rate of “<0.001%”. A hallucination is an output that looks plausible but is fabricated or wrong.

What the Attorney General found

The OAG’s investigation concluded those metrics were likely inaccurate and may have deceived hospitals about the accuracy and safety of the products, contrary to the DTPA.

The software had already been deployed at at least four Texas hospitals at the time.

Pieces denies any DTPA violation. The matter was resolved through an assurance of voluntary compliance rather than a finding of liability.

What was agreed

Notably, no monetary penalty was imposed. What Pieces accepted instead runs for five years and is arguably harder:

  • Any marketing that cites a metric, benchmark, or similar measurement of its generative AI outputs must clearly and conspicuously disclose both what the metric means and how it is calculated. Alternatively, Pieces may engage an independent third-party auditor to substantiate the claim.
  • No false, misleading, or unsubstantiated representations about accuracy, testing methodology, monitoring, or what its metrics mean.
  • No misrepresenting whether an endorser or reviewer is independent, and no omitting financial arrangements with endorsers.
  • Written disclosure to every current and future customer of any known or reasonably knowable harmful use or misuse, including risks of physical or financial injury from inaccurate output.
  • Documentation of the data and models used for training, the intended purpose, and known limitations.

The transferable point

The company was not punished for its software underperforming. It was pursued for publishing a precise-sounding accuracy number without disclosing how that number was produced.

A buyer’s practical defence follows directly: when a vendor quotes an accuracy figure, ask what it measures and how it was calculated. A vendor that cannot answer is making the exact claim Texas has already acted on.

What is not established

An assurance of voluntary compliance resolves an investigation without any admission or finding of wrongdoing, and Pieces expressly denies violating the DTPA.

We have not yet contacted Pieces Technologies for comment. When we do, the attempt and any response will appear on this entry.

Sources

Every claim above rests on one of these. Open them and check.

  1. Enforcement action September 18, 2024
    Attorney General Ken Paxton Reaches Settlement in First-of-its-Kind Healthcare Generative AI Investigation
  2. Published journalism October 10, 2024
    Texas Attorney General's Office Reaches Settlement with AI Company Over Deceptive Claims
  3. Published journalism September 20, 2024
    Novel Settlement Reached in Generative AI Deceptive Trade Practices Investigation

Right of reply

No contact has been logged for this entry yet.

If you are named here and something is wrong, dispute this entry. There is no charge, the route never expires, and your response is published unedited.

Other ai & automation entries

See all ai & automation entries, or browse the full record.


First recorded July 31, 2026 · Last updated July 31, 2026