A Texas company has begun selling corrections agencies an artificial intelligence system that listens to prison phone calls. It works out who is speaking, whether or not the speaker wants that.
LEO Technologies, based in Austin, announced Verus Voice AI on Monday. The product uses voice biometrics to catch prisoners who call on someone else’s personal identification number. It builds and tracks voice identities continuously, the company says, rather than waiting for staff to spot anomalies.
“Personal identification codes and phones can be shared, coerced, or swapped amongst inmates,” said Tim Anderson. “But a voice can’t be.” Anderson is a senior vice president at the firm.
Two descriptions of the same product
The press release stresses restraint. It says the system carries “built-in safeguards”, and that qualified personnel review every finding before anyone acts. The company promises “human oversight and accountability” throughout.
Its own product page makes a wider promise. The page invites agencies to “upload any voice sample and instantly search for matches across all conversations”. That, it says, eliminates manual verification.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
A companion page for the firm’s voice biometrics line is blunter. It reads: “Identify who’s speaking, even when they don’t want to be identified.”
Those are not descriptions of a PIN checker. They describe a searchable archive of voices, queryable against any sample an investigator supplies.
Fifteen years of voiceprints
Voice biometrics in American jails are not new. Securus Technologies has sold a product called Investigator Pro for over a decade. It markets the tool on nearly identical terms: detecting PIN abuse and recognising callers.
In 2019, The Appeal and The Intercept reported that institutions had built databases of hundreds of thousands of voiceprints. They found jails in Arizona, Texas, New Mexico, Michigan, Florida, Massachusetts, Colorado and Wisconsin buying the technology.
Much of that collection covered people awaiting trial. Officials at the Fort Bend County jail in Texas said the facility printed everyone who came through the door. A print then stayed in the system after release, unless an expunction order arrived. The vendor held the data on its own network, not the county’s.
Consent, in a place with very little of it
Civil liberties lawyers have long argued that such agreement means little. Aaron Mackey of the Electronic Frontier Foundation told The Appeal that jail voiceprinting could amount to illegal search and seizure. People submit, he said, because refusing costs them contact with their families.
Senator Ron Wyden, an Oregon Democrat, raised the commercial question. If Securus profits from biometric data the government seizes, he said, “Americans have a right to know”.
Clare Garvie, then at Georgetown Law’s Center on Privacy and Technology, went further. Agencies should destroy biometric templates, she argued, when prosecutors drop charges or a jury acquits. Several defence lawyers said nobody had told them the prints existed.
The other voice on the line
Every monitored call has at least two people on it. Only one is a prisoner. The EFF has noted that this class of technology could profile anyone whose voice crosses into a prison. That includes the parents, children and friends of the people inside.
LEO Technologies has not said whether Verus Voice AI enrols or keeps the voices of people outside the facility. Its announcement gives no retention period, no deletion process, and no storage location. Neither does the product page.
The gap matters because voice analysis rarely stops at identity. Meta has patented mood tracking from voice, and consumer tools now record conversations by default. Anthropic began collecting biometric data from flagged users this year.
The case for it
The problem the product targets is real. Phone minutes work as currency inside facilities, and prisoners trade, borrow and seize each other’s PINs. Investigators also use call monitoring to detect witness intimidation and the harassment of domestic violence victims. Securus cited exactly that in 2019 when it defended its own system.
LEO Technologies frames its platform around safety rather than punishment. The line includes video analysis, an operations dashboard, and a wellness tool for suicide prevention. Steven Harpe, its chief product officer, said earlier tools had been “too expensive and too unreliable”.
Accuracy remains the awkward part. The Appeal watched a Fort Bend sergeant review calls that Investigator Pro had flagged. The system rated its own matches at 83, 79, 55 and 45 percent. The sergeant accepted that defence lawyers could challenge such evidence, then sent the recordings to prosecutors anyway.
Already on the shelf
Verus Voice AI is on sale now. LEO Technologies holds a General Services Administration schedule contract and three Texas state purchasing contracts. Agencies can therefore buy it through existing routes, without a fresh procurement or a public debate.
That mechanism matters more than any single launch. It is how Flock’s camera network spread across American towns, and how facial recognition reached British high streets.
Bianca Tylek of the Corrections Accountability Project made the point in 2019, and it has aged well. Firms test this kind of system on people the public tends to treat as having forfeited their rights. “Today it’s this population,” she said, “but tomorrow it’s another.”
