Skip to content
Submit incident
AII-1002-55109
ReportedReported by fewer than three independent publishers, with no official finding.

ACLU warns against AI-generated police reports

The ACLU issued a white paper warning against police departments' use of AI-generated police reports, citing concerns about hallucination, bias, transparency, and accountability. Police departments in California, including Fresno, have reportedly begun using Draft One from Axon, which relies on GPT-4, to draft reports from body camera recordings.

10 December 2024 Β· OpenAI Β· Reliability and hallucination
10 December 2024
AII-1002-551092 sources ↓
Quick facts
Category
Reliability and hallucination
Country
United States
Source
Mashable
Date
10 December 2024
Drafted by
Raidu Incident Detection

What happened

Police departments in California are using Draft One from Axon to transcribe body camera recordings and create first drafts of police reports. The Fresno Police Department said it is using Draft One under a pilot program for misdemeanor reports, with the tool relying on OpenAI's GPT-4 model.

Reported impact

The ACLU warned that AI-generated police reports could produce errors that affect evidence and court cases, citing the technology's unreliability and bias. The group raised concerns that AI reports might omit facts, enable officers to lie about actions not captured on camera, remove accountability around discretionary power, and lack transparency in how they operate.

Index analysis

The ACLU identified four major concerns: AI systems are unreliable and prone to hallucination and bias; an officer's memories should be recorded before contamination by AI storytelling; the public and defendants lack transparency to understand how the systems work and interrogate the evidence; and the use of AI might remove accountability around discretionary power.

Governance controls

The ACLU stated it does not believe police departments should allow officers to use AI to generate draft police reports. The organization suggested that if AI tools are used, they should only transcribe audio-recorded verbal narratives created by officers, which could then be submitted alongside the original recording for review.

Classification

Category
Reliability and hallucination
Industry
Government, security and defence
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Unknown
Harm types
Human or fundamental rights, Public interest
Harmed
General public
Failure modes
Hallucination, Biased model
Consequences
None recorded
AI systems
Draft One

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

Mashable
Rising use of generative AI by police is a threat to Americans' civil liberties, ACLU warns
12 December 2024
engadget
ACLU highlights the rise of AI-generated police reports -- what could go wrong?
12 December 2024