Why You Shouldn't Use ChatGPT for Criminal Defense | Be My Own Attorney

Why Generic AI Fails

The Difference Between a Generic AI and One Configured for Your Case

A configured defense tool and a generic chatbot are not the same thing. The difference can determine your outcome.

ChatGPT, Gemini, Claude — the major consumer AI tools — are extraordinarily capable for most tasks. Criminal defense preparation is not one of them. Not because they are bad at reasoning, but because criminal defense is jurisdiction-specific, judge-specific, and case-specific. Generic AI is none of those things.

Join the Waitlist

Four Specific Failures

What Generic AI Gets Wrong

01

Hallucinated Case Citations

Large language models generate plausible-sounding case citations that do not exist. This is a well-documented failure mode — not a bug that will be patched, but a structural feature of how probabilistic text generation works. The model predicts what a real case citation looks like and produces one. If no real case fits the pattern, it invents one that does.

In a legal filing, citing a fabricated case is catastrophic. Courts have already sanctioned attorneys and pro se defendants who submitted AI-generated briefs with invented citations. The judge who reads your motion and finds a hallucinated case will dismiss your argument and your credibility simultaneously. Brief uses verified case law databases — not generated case citations. Every citation it produces can be checked.

02

No Jurisdiction Knowledge

Criminal procedure is jurisdiction-specific. How to file a suppression motion in Texas is different from California is different from Ohio. Local court rules vary by county, by courthouse, sometimes by courtroom. Filing deadlines differ. Form requirements differ. What constitutes a timely objection differs.

Generic AI gives you the general principle, not the local rule. "You can file a motion to suppress" is not the same as knowing that Harris County district courts require motions to be filed in a specific format with specific notice requirements. Brief is built with jurisdiction-specific procedures loaded — the rules that actually govern your case, not the national average.

03

No Judge Intelligence

A generic AI has never heard of your judge. It cannot tell you that this particular JP judge has been reversed on Fourth Amendment suppression twice in the last three years. It cannot tell you that she hates speaking objections or that she has never sustained a motion to dismiss from a pro se defendant. It cannot tell you what arguments have worked in front of her before.

Brief's judge database is built from public records: PACER, state court records systems, FOIA responses, news archives, and appeals data. It knows your judge's actual ruling patterns on the issues relevant to your case. That is not something any general-purpose AI tool can replicate without that specific data.

04

No Case-Specific Analysis

Generic AI analyzes a hypothetical version of your situation. When you describe your case to ChatGPT, you are giving it a summary from memory. The AI responds to that summary — not to your actual police report, not to your actual discovery documents, not to the specific language in the charges filed, not to the actual evidence the prosecution has.

Brief analyzes your actual documents. The suppression analysis comes from reading your specific police report, not from a description of it. The charge analysis comes from the actual charging instrument, not a paraphrase. The cross-examination questions come from the witness statements in your discovery, not a hypothetical witness. The gap between analyzing a description of a case and analyzing the case itself is the gap between preparation and the illusion of it.

Evidentiary Risk

Public AI Tools and Your Case Documents

When you paste your police report, your charges, or your case strategy into a public chatbot, you are transmitting sensitive case information through a third-party service with its own terms of service, data handling practices, and potential evidentiary exposure.

Courts have begun examining what happens to communications a defendant makes through third-party AI tools in terms of privilege. For represented defendants, this creates attorney-client privilege complications. For pro se defendants, the analysis is different — but the evidentiary exposure from transmitting case details through a public commercial service is real regardless of privilege status.

Brief processes your case documents through a private AI instance with contractual data protections. Your prompts do not train any model. Your documents are not shared with the underlying AI provider for any purpose. We designed specifically for the use case where the documents being analyzed are the kind you would not want in a commercial provider's training data. Read our full data privacy policy →

What We Do Instead

What Configured Looks Like

Brief is not a chatbot you describe your case to. It is a configured analysis system that reads your actual case documents and queries a live database of verified legal information. Here is what that means in practice:

Private Instance

Your documents run through a private API instance, not a public consumer product. Contractual data protections apply. Your information does not enter a shared training pool.

Case-Specific Context

Your actual documents — police report, charges, discovery — are ingested. Analysis is based on what's actually in your case, not a description of it from memory.

Jurisdiction-Specific Case Law

Verified case law and local rules for your specific jurisdiction. Not national averages. Not generalized principles. The law that actually governs your court.

Live Judge Database

Built from PACER, state court records, FOIA responses, and appeals data. Your judge's actual ruling patterns — not a generic judicial profile.

The Prior Attempt

The Attempt That Already Failed — and Why Ours Is Different

In early 2023, DoNotPay announced that a defendant would wear an AI earpiece in court and be coached in real time. Bar associations in multiple states threatened unauthorized practice of law charges. The plan was abandoned before it was tested.

The important detail: the threat came to the company's CEO, not to the defendant. The UPL (unauthorized practice of law) threat was directed at DoNotPay as a company providing legal services to clients — the same legal theory used against non-attorney legal document preparers, legal tech companies, and anyone outside the bar who provides legal services to others.

The DoNotPay model was structurally vulnerable because it was framed as a company providing legal services. A defendant using tools to prepare and conduct their own defense occupies a constitutionally protected position that the DoNotPay model did not. Your right to conduct your own legal research, to prepare your own defense, and to use available tools in doing so has never been the target of UPL enforcement — because UPL prohibits non-attorneys from providing legal services to others, not defendants preparing their own cases.

We are a technology company that builds tools. You are a pro se defendant using those tools to prepare your own defense. That structure is constitutionally different from a company providing legal services to clients. We built this distinction in from the beginning. Read the legal foundation →

Questions

Frequently Asked Questions

You can use any research tools you want. The concern with using generic AI for legal research isn't that it's prohibited — it's that it produces unreliable results in jurisdiction-specific, case-specific contexts. If you use ChatGPT for general research, verify everything independently before relying on it in a filing or a proceeding. Never cite a case you haven't verified exists and says what you think it says.
No. Brief uses a private AI instance — a direct API integration with contractual data protections — not a consumer product like ChatGPT, Gemini, or any other public-facing tool. Your documents are processed in a private context. They do not pass through a consumer AI product's data pipeline.
We use frontier AI models through private API access. We don't disclose the specific model because the important distinction isn't which underlying model is used — it's the private instance configuration, the case-specific context loading, the verified legal databases, and the judge intelligence data that separate Brief from typing into a consumer chatbot.
The primary method is grounding. Brief's case law citations are pulled from verified legal databases and cross-referenced — the system is not generating case names, it is retrieving real cases. Judge intelligence comes from court records, not generation. Jurisdiction-specific rules come from official court documents, not inferred from training data. No system eliminates hallucination entirely — which is why we are explicit that all citations should be independently verified before reliance. We make it easy to verify because we tell you exactly where each piece of information comes from.
Verify before you rely. This is not a hedge — it is the fundamental operating principle for any legal research tool, human or AI. Brief provides sourcing for its analysis. Verify case citations exist and say what Brief says they say. Check local rules against the court's official website. Use Brief to identify what to research and where to look — then verify the findings before you act on them. This is how competent legal research works, regardless of who or what generates the initial analysis.

Built Different

Join the Waitlist

Brief launches in Texas first. Configured defense intelligence — your actual case, your actual judge, your actual jurisdiction — not a generic chatbot. Learn more about Brief → or Stand → for real-time courtroom coaching.

Join the Waitlist

Not legal advice. Not an attorney-client relationship. Verify all citations before reliance. See our privacy page and disclaimer.