Code of Ethics

The legal knowledge required of a modern police officer is staggering. Officers must understand and apply state criminal statutes, codes of criminal procedure, constitutional law from the Fourth Amendment through the Fourteenth, Supreme Court precedent, federal circuit rulings, state appellate decisions, departmental policy, and DA guidelines — often in real time, under stress, with incomplete information. No human being can carry all of that in their head.

The current alternatives are inadequate. Officers ask coworkers who may be wrong. They post questions on social media forums where bad advice is indistinguishable from good. They spend hours manually searching case law databases designed for attorneys, not for officers who need an answer before their shift ends. Or they use general-purpose AI tools that hallucinate citations, fabricate case law, and present fiction with the confidence of fact — the same tools that have already gotten attorneys sanctioned by courts for submitting fake cases.

LawCite was built by an active police officer who lived this problem. Not by someone who studied it from the outside. Not by a technology company looking for a market. By a cop who watched his colleagues spend hours doing what should take seconds, and who watched the consequences when they got bad information instead.

We built LawCite on Anthropic's Claude because of its Constitutional AI methodology. This was not a convenience decision. It was a values decision.

Constitutional AI is a framework for training AI systems that are helpful, honest, and harmless — using a set of explicit principles to guide the model's behavior rather than relying solely on human feedback, which carries its own biases and limitations. For a tool that will be used by law enforcement officers making decisions that affect people's liberty, we believe the underlying architecture of the AI is not a technical detail. It is a moral one.

We then layer our own safeguards on top: hallucination prevention systems that catch fabricated information before it reaches the user, citation verification that confirms every case reference actually exists and says what the AI claims it says, and anti-bias prompting that guards against the system drifting toward telling users what they want to hear.

We do not trust any AI system to be right by default. We verify. And then we verify the verification.

Tell the truth no matter what, no matter how bad it makes you look.

LawCite exists to surface what the law actually says — not what the user hopes it says. We will tell an officer when the law does not support their position. We will surface case law that weakens their argument alongside case law that strengthens it. We will recommend they contact their legal department or district attorney when the analysis reveals a problem that needs to be remediated, not rationalized.

This is not a popular feature. Some agencies will push back when our tool tells their officers they were wrong. We accept that. If an agency believes the analysis is incorrect, they should be able to demonstrate it with facts the AI did not have access to. The AI can only work with the facts and context it is given, and it can be wrong. But it will never be dishonest.

We would rather lose a customer than help someone lie.

LawCite is a tool for law enforcement, but our allegiance is to justice itself. These two things should be aligned — and when law enforcement operates as it should, they are. But we refuse to pretend that a tool built for officers is automatically a tool that serves justice. It only serves justice if it is built that way on purpose.

That means we operate with fairness toward all parties in the legal system. The suspect. The officer. The victim. The prosecutor. The defense attorney. The public. Our analysis does not take sides. It takes the law seriously.

This is a tool to help officers be better — not to help anyone cover up mistakes, build retroactive justifications, or manipulate legal reasoning to reach a predetermined conclusion. LawCite is a tool for preparation and analysis, not for retroactive justification.

Every sworn officer knows these words. They are the bedrock of the Law Enforcement Code of Ethics — the promise that an officer's fundamental duty is to serve the community, to uphold the Constitution, and to honor the rights of all to life, liberty, equality, and justice. LawCite is built on the same oath.

Duty means an officer at 2 AM on a traffic stop can get a clear, accurate legal analysis in seconds instead of hours. It means a detective building a case can verify their understanding of the relevant statutes and case law without waiting for a legal advisor who may not be available. It means a command staff member can make policy decisions informed by the actual state of the law. Our duty is to make their duty easier to fulfill.

Integrity means LawCite tells the truth even when the truth is not the answer the user wants. It means the system cannot be sweet-talked, led, or manipulated into agreeing with a flawed legal position. It gives the same reasoning and output based on the facts — regardless of whether the user is an officer trying to find a way out of a mistake or a supervisor trying to get an officer in trouble when they did nothing wrong. Just as an officer must never allow bias or personal gain to influence their decisions, LawCite must never allow user pressure to influence its analysis.

Accountability means LawCite provides legal analysis, not operational directives. We will analyze the legal framework around use of force — we will not tell an officer what level of force to use. We will analyze the legal basis for an arrest — we will not tell an officer whether to make one. The distinction is between “here is what the law says and how courts have ruled” and “here is what you should do right now.” We draw that line deliberately, because the consequences of blurring it are measured in lives. The officer remains accountable for their decisions. LawCite remains accountable for the accuracy of its analysis.

Good policing is practical. You learn fast that the best solution to a problem is usually the most direct one — the one you can execute reliably under pressure, not the one that looks impressive on paper. We build LawCite the same way. If it works and it's reliable, we ship it. If it's clever but fragile, we don't.

But simplicity is not the same as carelessness. We spent months building proper authentication and multiple methods of hallucination elimination rather than slapping a new interface over an existing AI model and calling it a product. The simple thing that works is often the thing that was built carefully, tested thoroughly, and designed to hold up when it matters most — not the thing that was thrown together fast.

When an officer relies on LawCite in the field, “it usually works” is not good enough. It has to work. Simplicity in design serves reliability in practice.

Bias is the quiet enemy of justice. In the context of LawCite, we are concerned with two specific forms of it.

Legal interpretation bias — the tendency to read the law in the direction that supports a predetermined conclusion. LawCite is built to resist this. Our analysis presents the law as it stands, including the tensions, ambiguities, and unfavorable precedents that a biased reading would conveniently omit.

Confirmation bias — the tendency of a user to seek out the answer they already believe is correct, and to pressure the tool until it agrees. LawCite is built to resist this too. The system should give the same output regardless of how the question is framed, how many times it is asked, or how much the user attempts to lead it toward a preferred conclusion. An officer looking for validation and an officer looking for truth should receive the same analysis — because the law does not change based on who is asking.

We implement anti-bias safeguards not because we distrust officers, but because we respect the weight of the decisions they make. Bias is a human tendency. Guarding against it is a professional discipline.

Everything we build, every decision we make, every feature we ship or decline to ship — all of it is measured against a single question: does this help law enforcement officers do their jobs better, in a manner consistent with justice?

If it does, we build it. If it doesn't, we don't. If it helps officers but undermines justice, we don't. If it sounds impressive but doesn't make an officer's shift measurably better, we don't.

The mission is not to build the most advanced AI product. The mission is not to raise the most capital or acquire the most customers. The mission is to give law enforcement officers quick, reliable access to accurate legal information so they can make better decisions, serve their communities with greater confidence, and uphold the law they swore to protect.

That mission comes first. Everything else is in service to it.

Defining what we refuse to do is as important as defining what we do.

LawCite will not make arrest decisions for officers. It will analyze the legal framework and relevant case law, but the decision to arrest is a human judgment that must remain with the officer on scene, who has access to facts and context that no AI system can fully replicate.

LawCite will not prescribe use-of-force actions. It will provide analysis of use-of-force law, standards, and relevant court decisions. It will not tell an officer what to do in the moment. That line exists because the consequences of crossing it are irreversible.

LawCite will not serve as a tool for retroactive justification. It is designed for preparation, analysis, and legal understanding — not for constructing a narrative after the fact. Officers and agencies that attempt to use it this way will find that the system resists it, because it is built to present the full picture, including the parts that are inconvenient.

LawCite will not allow itself to be manipulated. The system is designed so that the same set of facts produces the same analysis regardless of how the question is framed, who is asking, or what outcome they are hoping for.

There is an inherent tension in building a tool for law enforcement that is also committed to justice above all else. We accept this tension. We do not hide from it.

There will be moments when LawCite's analysis is not what the officer wants to hear. There will be moments when an agency questions why our tool told their officer that the law does not support their position. There will be moments when the commercially easy thing would be to soften the analysis, hedge the conclusion, or let the user lead the AI toward a more comfortable answer.

We will not do this. Not because we are adversarial toward law enforcement — we are law enforcement — but because the entire value of this tool depends on its integrity. The moment LawCite starts telling people what they want to hear instead of what the law actually says, it becomes worthless. Worse than worthless — it becomes dangerous.

An officer who trusts LawCite must be able to trust it completely. That trust is only possible if the system is honest even when honesty is uncomfortable. Especially when honesty is uncomfortable.

LawCite will grow. We plan to allow agencies to integrate their own policy manuals, creating a system that can analyze not just the law but the specific policies and procedures of the officer's own department. We plan to expand our coverage of jurisdictions, statutes, and case law. We plan to build features we haven't imagined yet.

But we will not grow at the expense of our values. Every new feature, every expansion, every partnership will be measured against the same principles laid out in this document. If we cannot build something without compromising the truth, we will not build it. If we cannot serve a customer without bending our commitment to impartial analysis, we will not serve them.

We would rather be a small company that officers trust absolutely than a large company that officers trust conditionally.

The fabric of a just society depends on the people who enforce its laws doing so with knowledge, integrity, and accountability. LawCite exists to make that easier — not by replacing the officer's judgment, but by ensuring that judgment is informed by the best available understanding of the law.

We are not building technology for the sake of technology. We are building a tool that we believe makes justice more accessible, officers more effective, and the legal system more fair. Every decision we make — from the architecture of our AI to the values on this page — is in service to that belief.

The law belongs to everyone. Understanding it shouldn't take hours. And the tools that help officers apply it should be as honest, impartial, and accountable as we expect the officers themselves to be.