AI in Criminal Defense: What Works and What Stays Human

Criminal defense attorneys are using AI to process body camera footage, tear through discovery in federal cases, and draft motions faster than any associate ever could. The adoption is uneven — large federal defense practices are ahead, solo defenders are catching up with consumer tools — but the efficiency gap between AI-equipped and non-AI criminal defense is already measurable in hours per case.

How AI Is Used in Criminal Defense Today

The biggest shift in criminal defense AI is happening in discovery. Federal criminal cases routinely produce hundreds of thousands of documents, and firms running AI-assisted review are cutting document review time by 40-60%. Relativity's AI tools flag relevant documents across massive discovery sets, identifying patterns in financial records, communications, and transaction data that manual review misses or takes weeks to find.

Body camera and surveillance video analysis is the second major use case. AI tools now timestamp, transcribe, and flag key moments in hours of footage — work that previously required paralegals to watch in real time. For a DUI defense, that means finding the 47-second window where the officer deviated from protocol without sitting through 3 hours of dashcam footage.

Sentencing analysis is where AI gets tactical. Tools like Lex Machina provide judge-specific analytics: average sentences by charge type, deviation from guidelines, receptiveness to particular arguments. Criminal defense attorneys are using this data to build sentencing memoranda grounded in actual judicial behavior, not guesswork.

Brief and motion drafting with Claude and ChatGPT is now standard among tech-forward defenders. The workflow isn't "AI writes the brief" — it's "AI produces a research-backed first draft in 90 minutes, attorney spends 2 hours refining it." That's a 70% reduction in total drafting time for routine motions to suppress or dismiss.

Best Tasks for AI in Criminal Defense

Discovery review is the highest-value AI task in criminal defense. The work is repetitive, the documents are structured, and the volume in federal cases is crushing. AI doesn't just speed up review — it catches patterns across thousands of documents that humans miss because they can't hold that much context. Relativity and similar e-discovery platforms handle this well for large cases. For smaller cases, feeding key documents into Claude or NotebookLM for synthesis gives solo practitioners BigLaw-level analysis capacity.

Case law research for novel defense theories is the second sweet spot. Criminal defense often requires creative arguments — constitutional challenges, procedural defects, novel interpretations of statutes. Claude and ChatGPT are strong at surveying broad legal landscapes and identifying analogous cases across jurisdictions. The key is using AI as the first research pass, then verifying every citation manually. NotebookLM is particularly useful here: upload the prosecution's filings, relevant statutes, and key cases, then have it synthesize connections.

Jury research and selection data aggregation is the third task where AI adds clear value. Compiling demographic data, social media analysis summaries, and historical voir dire patterns into structured profiles saves hours of prep time before jury selection. The attorney still makes the strike decisions — AI just ensures they walk in with better information.

What Stays Human

Courtroom advocacy stays human. Full stop. Cross-examination is improvisation — reading a witness's hesitation, adjusting tone, exploiting an inconsistency in real time. No AI handles this because it requires reading micro-expressions, sensing jury reactions, and making split-second strategic pivots based on what just happened 4 seconds ago. The best cross-examiners are performers, and performance isn't delegatable.

Client counseling in criminal defense carries emotional weight that AI can't touch. When you're advising someone facing 15 years on whether to take a plea, that conversation requires trust built over months, understanding of the client's family situation, and the ability to deliver devastating news with empathy. Criminal defense clients are often in custody, scared, and distrustful of the system. The human relationship IS the representation.

Plea negotiation strategy depends on relationships with prosecutors and judges that take years to build. Knowing that this particular ADA will move on a reduced charge if you present the right mitigating facts, or that this judge responds poorly to aggressive motions — that's institutional knowledge and political awareness that AI doesn't have and can't develop. The same applies to sentencing advocacy: standing before a judge and humanizing your client requires presence, credibility, and emotional intelligence that no model replicates.

Tools and Workflows That Work

For research and drafting, Claude is the strongest option for criminal defense. Its long-context window handles lengthy case files, and its writing quality produces motion drafts that need refinement, not rewrites. ChatGPT works well for interactive strategy sessions — explaining a case theory and having it stress-test weaknesses. Use both: ChatGPT for brainstorming defense theories, Claude for the actual drafting.

For case material synthesis, NotebookLM is underrated. Upload the prosecution's discovery index, key witness statements, and relevant case law — then query it for connections and contradictions. For large federal cases with serious discovery volume, Relativity is the standard e-discovery platform, and its AI features are worth the cost at that scale. For judge analytics, Lex Machina provides sentencing patterns, motion success rates, and judicial tendencies that inform strategy.

Here's the practical workflow: Intake comes in, you run the case facts through Claude for a preliminary legal issues assessment. You pull judge analytics from Lex Machina. You use NotebookLM to synthesize the discovery materials. You draft motions in Claude with the synthesized research as context. You review, refine, and file. This workflow doesn't require expensive legal AI wrappers — it's general-purpose AI tools configured for criminal defense work. Build the system around the model, don't pay $500/seat/month for someone else's wrapper.

Disclosure and Compliance

Criminal defense faces the most uncomfortable AI disclosure dynamic in legal practice. Multiple federal district courts now require attorneys to certify whether AI was used in preparing filings. For defense attorneys, this creates a strategic tension: disclosing AI use in a brief could signal to prosecutors or judges that the defense is under-resourced, or it could invite heightened scrutiny of the work product. Not disclosing risks sanctions if the court requires it.

The practical answer is compliance with whatever the jurisdiction requires, documented internally. If you're in the Southern District of New York, you certify. If your state court hasn't issued an order, you still need a defensible internal policy because it's coming. The ABA's Formal Opinion 512 (2024) makes clear that attorneys must supervise AI use and take responsibility for AI-assisted work product — the standard of competence doesn't change because a model helped draft the brief.

The unique criminal defense risk is fabricated case citations. The stakes are higher here than in civil practice — a motion to suppress based on a hallucinated case could result in ineffective assistance of counsel claims, sanctions, and bar discipline. Every AI-generated citation must be verified in Westlaw or Lexis before filing. No exceptions. Build verification into the workflow as a mandatory step, not an afterthought.

The Bottom Line

Criminal defense AI adoption is early but accelerating fast, especially in discovery-heavy federal cases and research-intensive motion practice. Start with one workflow — discovery review or motion drafting — and build a verification protocol before scaling.