A U.S. telecom company, Securus Technologies, has developed an artificial intelligence model trained on years of prison phone calls to predict and prevent crimes among inmates. The company reports that it began building these AI tools in 2023, utilizing a vast database of recorded calls to create models tailored to specific states and counties. Securus has been piloting these tools to monitor inmate communications in real time, aiming to detect potential criminal activities early. While the company claims that its AI system has helped disrupt human trafficking and gang operations, advocates for prisoners’ rights express concerns about civil liberties violations. Bianca Tylek, executive director of the advocacy group Worth Rises, argues that inmates are not fully informed that their recorded conversations will be used to train AI models, labeling this as “coercive consent.” Additionally, recent Federal Communications Commission reforms have changed how telecom companies like Securus can use funds from inmate calls, prompting discussions about the financial implications for both the companies and the incarcerated individuals.
Why do we care?
This is a preview of where surveillance AI is heading, and it should raise red flags. You’ve got a company training models on years of recorded calls — conversations inmates had no meaningful way to opt out of — and then using that to make predictions with real consequences. That’s not a standard you want anywhere near your clients.
Because what happens next? These tools get repackaged. Suddenly it’s “employee monitoring” or “insider threat detection,” and now your customers want to analyze Teams calls and Slack messages with the same level of scrutiny. And you’re the one they’ll ask to implement it. If you don’t have a governance framework, you’re walking straight into liability.
The big risk here is false positives. If an AI model misinterprets a call in a prison, someone gets punished. If it misinterprets a conversation in a business, it can tank careers or trigger HR actions based on bad data. And none of these systems come with the auditability or accuracy metrics you’d need to trust them.
So this is a moment to get ahead of it. Start drawing clear lines with clients about what data can be used for AI training. Build disclosure templates. Push for transparency and consent. And when someone shows up with a behavioral prediction tool, don’t just plug it in — ask about bias, error rates, and governance. This is the kind of technology that can creep into the workplace quietly. Your job is to put the brakes on until the guardrails are in place.