Did you find this evidence or did your AI?

Meanwhile, in Dubai, real time video feeds run through Oyoon, a computer vision system that can identify plates, track vehicles, or flag unusual behaviour patterns at a scale a human operator could never match. In London, the Metropolitan Police use live facial recognition for specific operations, generating alerts that influence tactical decisions.

Across the UK, investigators use tools such as altia Transcribe, a specialist law enforcement transcription tool that turns hours of interviews or surveillance audio into near instant structured text. It tags speakers, timestamps every segment, and creates an audit trail. The tool is designed so that the transcript can flow directly into case management systems and the evidential chain.

And then there is Axon Draft One, already deployed in some forces, which uses generative AI to convert body worn camera footage into first draft police reports. The narrative of an encounter, once written entirely by an officer, can now begin life inside an algorithm

AI is no longer a “future question” in criminal justice. It is already an invisible layer woven through investigations, disclosure, and courtroom preparation.

When the evidence itself is shaped by AI

In courtrooms, challenges are emerging around evidence that has been processed, enhanced, or generated with AI assistance.

In a Washington case, a court rejected AI enhanced video because the generative method used did not meet accepted forensic standards.

In Wisconsin’s State v. Loomis, a risk scoring algorithm was permitted, but only with explicit warnings about its limitations and with restrictions on how it could influence sentencing.

The principle behind these decisions is not new. Courts require that any analytical technique must be explainable, validated, and capable of being challenged. The difference now is that many tools do not simply analyse evidence. They help decide what evidence investigators focus on, and what prosecutors later disclose.

Large studies on facial recognition performance, including those conducted by NIST, show that certain demographic groups can experience different false match rates depending on the system. Forensic AI researchers are responding by building “glass box” models designed to be interpretable enough to explain in a courtroom how a conclusion was reached.

The deluge and the rise of AI as an evidence gatekeeper

A major criminal investigation can now generate millions of digital documents, thousands of hours of video, and vast social media archives. Manual review is no longer possible. That is why tools like Longeye and Altia Transcribe have become central not only to investigations, but to disclosure obligations.

These systems ingest, cluster, and categorise material, highlight patterns, and flag “likely relevant” content. When an investigator uses these tools, the path the AI took to reach those conclusions becomes part of the evidential story. The audit trail becomes as important as the files themselves.

Regulatory frameworks are adapting to this reality. The EU AI Act categorises many policing tools as high risk, requiring extensive logging of how outputs were generated, mandatory oversight, and strict documentation. This aligns with long standing forensic expectations: version control, validation studies, and error rate evidence are essential.

The justice system is quietly shifting from “show me the evidence” to “show me how your AI found this evidence.”

AI in the courtroom: transcripts, translation, and access

Inside the courtroom, AI is increasingly used to improve accessibility. The UK Supreme Court has been testing tools that combine automatic speech recognition with large language models so that hearings can be rapidly transcribed, tagged by speaker, and linked to passages in the final judgment.

The University of Surrey has adapted AI models specifically to courtroom speech, overcoming issues like overlapping dialogue and specialised legal terminology. Altia Transcribe is already being used to prepare transcripts that integrate directly into case bundles.

Courts are also exploring AI supported translation to meet language access duties. Guidance warns that while AI can help provide rapid drafts, human interpreters must check accuracy where rights are affected.

These tools do not replace the official record, but they allow lawyers, defendants, and the public to navigate otherwise overwhelming volumes of spoken and digital evidence.

What justice systems are doing to prepare

Across jurisdictions and institutions, several concrete steps are emerging.

Accountability, testing and human supervision – First, governance frameworks now treat AI as part of justice infrastructure. The UK Justice Ministry’s AI Action Plan and Europol’s AI and policing strategy emphasise accountability, testing, and human supervision. Prosecutors such as the Crown Prosecution Service publish explicit statements about the limits of AI use in casework.

Explainable with an audit trial – Second, technical standards are being built around validation, audit trails, and explainable. Forensic AI research stresses that tools must be tested on real world data, with documented error rates and clear logic. The EU AI Act hard codes logging and oversight obligations.

Training, training, training – Third, training for judges, defence lawyers, and investigators is becoming essential. International bodies note that legal professionals do not need to understand coding, but must know how to question algorithmic outputs.

Access for defence – Fourth, defence access to model documentation, validation studies, and expert support is increasingly recognised as necessary to ensure a fair trial.

Across all of this lies a shared assumption: AI tools must be treated like forensic instruments. Their outputs, limitations, and decision processes must be transparent and challengeable.

Where this leaves us

AI now touches almost every stage of the criminal process. It extracts leads, sorts evidence, drafts reports, transcribes hearings, and helps courts manage information overload. None of this removes human judgment. It does, however, demand new scrutiny.

The emerging challenge is not whether AI should be used, but how its role is explained and defended when evidence enters court. Soon enough, a barrister will rise and ask an officer:

“Did you find this evidence yourself, or did your AI?”

And the system must be ready to answer that question, clearly and fully.

AI at Altia

Altia supports AI-enhanced investigations and OSINT by helping investigators handle large volumes of digital material quickly while maintaining evidential integrity. Tools such as Altia OSINT Investigator enable investigators to ingest, structure, and analyse open-source data, audio, video, and online content, surfacing relevant connections while preserving full audit trails and traceability.

The Altia OSINT Investigator is designed for modern investigative workflows, combining AI-assisted discovery with transparency, logging, and human oversight. This ensures that insights generated from OSINT can be explained, disclosed, and defended in court, aligning with emerging legal and regulatory expectations around explainable and accountable AI in criminal justice.