In partnership with
Keep up with tech in 5 minutes
TLDR is the free daily email with summaries of the most interesting stories in startups, tech, and programming. The stuff worth knowing, minus the doomscrolling.
Issues are curated by ex-Google and Anthropic engineers and land in your inbox before your morning coffee. A 5-minute read, and you walk into the day already knowing what your team is still catching up on.
Tech is just the start. We also cover AI, marketing, dev, and more. Pick the briefs that match your work.
Free, daily, and read by 7M+ subscribers. Subscribe and let the experts do the digging for the tech news that matters.
🚨 An investigator sits across from an officer in an internal affairs interview. The recorder runs.
Weeks later, that same audio gets fed into a piece of software. The software scores the officer's statements for deception markers: tone, pacing, micro-hesitation. It flags portions as "high probability untruthful."
No polygraph. No stipulation. No cross-examination of the method. Just a score, sitting in a file, waiting for someone to ask where it came from.
That scenario isn't science fiction. It's the direction internal affairs work is heading right now. Most agencies don't have a policy for it yet.
What Happened
The first wave of AI in policing was about words on a page. Generative tools turned body-camera audio into report narratives. IA commanders debated it. Some agencies banned it. Most moved cautiously.
That conversation is already outdated.
AI is no longer just drafting reports. It's summarizing case files for supervisors. It's correlating dispatch records with body-camera footage. It's flagging patterns across jail calls and surveillance video. And, most consequentially, it's scoring human truthfulness.
Each of those functions touches the same thing IA exists to protect: the integrity of the finding.
Here's the real question every commander needs to sit with: if an investigator can't fully explain how a tool reached its output, can they fully defend the finding built on top of it? That's not a technology problem. That's a due process problem. And due process is the entire foundation IA legitimacy rests on.
What Is AI Actually Doing Inside Investigations Right Now?
Case correlation and briefing
Vendors like Mark43 now offer tools that go beyond report drafting. ReportAI helps officers draft reports using information from dispatch records and body camera footage, while BriefAI summarizes case information for investigators and supervisors. Agencies can choose which AI features to enable and who can access them, and the system maintains audit logs of AI-assisted activity.
That audit trail matters. It's the difference between a tool IA can defend and one it can't.
Pattern And Lead Generation
Departments are now using AI to analyze jail calls and surveillance video, and to identify vehicles through distinctive features beyond license plates. This sits alongside real-time crime center technology, predictive analytics, and live translation tools. Advocates frame these tools as a force multiplier, not a replacement for human judgment, investigative rigor, or community trust.
That framing is correct. It's also aspirational until an agency writes it into policy.
Behavioral and deception analysis
This is the frontier IA commanders should watch most closely.
Academic and government-funded research has produced AI models that analyze facial expressions, speech patterns, and body movements to assess truthfulness. Some studies claim accuracy over 90% in various scenarios. RAND has tested machine-learning methods for detecting deceptive speech patterns in simulated background interviews.
These tools are not polygraphs. But they carry the same core risk polygraphs carried for decades: a confident-looking score that outruns its own scientific footing, applied to a human being whose career is on the line.
Transparency and audibility
The most important cautionary tale so far isn't about accuracy. It's about erasure.
Electronic Frontier Foundation investigators found that Axon's Draft One, the leading AI report-writing tool, seems designed to stymie any attempts at auditing, transparency, and accountability. Once an officer edits and exports a draft, the tool erases the initial draft. With it goes any evidence of what portions of the report were written by AI and what portions were written by an officer.
The King County Prosecuting Attorney's Office in Washington responded by barring the practice outright. Their stated position: "For now, our office has made the decision not to accept any police narratives that were produced with the assistance of AI."
The regulatory ground is shifting too
The federal government has established a DOJ AI Litigation Task Force aimed at challenging state AI laws nationwide on constitutional and preemption grounds, part of a push toward a federal policy of "minimally burdensome" AI regulation.
Whatever position an agency takes today may need to be revisited as this fight plays out. Governance built on today's state law could be built on sand.
What Should IA Commanders Do Right Now?
This isn't a call to ban AI. It's a call to govern it before it governs you.
Treat AI outputs as evidence, not conclusions. A deception score or an AI-generated summary is an investigative lead, never a finding. The investigator's independent judgment, documented in the file, remains the basis for any disciplinary outcome.
Preserve everything. If a tool doesn't retain original outputs, prior drafts, and a clear record of what the AI generated versus what a human wrote or edited, that tool is a liability, not an efficiency gain. Insist on audit logging as a condition of procurement.
Write the policy before the vendor writes it for you. Define which AI functions are permitted, which require supervisor sign-off, and which are prohibited outright, particularly anything touching credibility assessment.
Train investigators on the tools, not just the topic. Investigators need to understand what a model can and can't validly claim, so they can explain and defend its use under cross-examination or grievance review.
Assume discovery. Opposing counsel, union representatives, and courts will ask what AI tools touched an investigation. If you can't answer precisely, you've created exposure that has nothing to do with the underlying misconduct.
Questions IA Commanders Should Ask
Do we know every AI tool currently touching an investigative file, including report writing, transcription, analytics, and behavioral scoring, across our agency?
If challenged in arbitration or court, can we show exactly what a human investigator did versus what a machine generated?
Do we have a written policy governing AI use in interviews and case analysis, or are we relying on vendor defaults?
Would our current practices survive the same scrutiny that got Draft One flagged by outside investigators?
Who in our chain of command is accountable if an AI-assisted finding is later shown to be wrong?
Leadership Takeaway
AI isn't the threat to internal affairs. Ungoverned AI is.
The agencies that come out of this decade with their credibility intact won't be the ones that adopted the most AI, or the ones that banned it entirely. They'll be the ones that built policy, audit, and accountability around it before the first grievance forced them to.
Public trust in internal affairs was never built on technology. It was built on the discipline to explain every decision, every time. That discipline doesn't change because the tool in the room got smarter. It becomes more important than ever.
If your agency is building its AI governance framework from scratch, or wanting to learn more about AI and its impact on internal affairs investigations, click here to learn more.
FAQ Schema Q&As
Q: Is AI deception detection technology reliable enough to use in internal affairs investigations? A: Research models claim high accuracy in controlled studies, but these tools have not been validated for operational, disciplinary, or courtroom use. Agencies should treat any AI credibility assessment as a lead, not a finding.
Q: Why did a prosecutor's office ban AI-written police reports? A: The King County Prosecuting Attorney's Office cited reliability and transparency concerns, noting AI-generated narratives could not yet be relied upon in cases affecting people's freedom.
Q: What should an internal affairs policy on AI include? A: At minimum, it should define permitted and prohibited AI uses, require preservation of original AI outputs and audit logs, and mandate independent investigator corroboration of any AI-generated conclusion.
