10 AI agent prompts for Legal & Compliance
Copy-paste, beginner-friendly prompts to help you do first-pass contract reviews and assemble compliance evidence. Each one is based on a real-world example and works with ChatGPT, Claude or an AI agent. 7 are free below, and 3 more are in AgentPlaybook Pro.
Free Legal & Compliance prompts
Compliance evidence pack
Gather documentation for a review, like Deutsche Bank's third-party-risk agent.
I need to prepare a [vendor/security/regulatory] review for [vendor or process]. List the evidence needed, check these documents [paste] against it, and show what's missing in a table.
Claim intake assistant
Collect complete claim info up front, like Allianz/Travelers claim agents (humans approve payouts).
Design a claim-intake conversation for [type of claim]. List required info, questions to ask in order, documents to request, and a summary format for the human who approves the payout.
Keep a running log of a contract negotiation
Have an agent capture what changed, why, and what's still open each time a new redline comes back, so negotiation context doesn't live in one person's head, like the agent Jeremy (@JHurst12) built while running an autonomous business at Swan AI.
You keep the negotiation log for our contract with [counterparty] for [deal/agreement]. Here is the previous version: [paste or attach]. Here is the new redline they sent: [paste or attach]. Here are the related emails and call notes: [paste]. Update the log: 1) list every clause that changed, with old wording, new wording, and who proposed it; 2) for each change, note the reason given (quote the email or note) or write 'reason not stated'; 3) list the points still open, what each side wants, and our fallback position if I've told you one ([paste our positions]); 4) flag anything new that wasn't discussed before. Keep the log in one table that grows with each round. Don't send anything to the other side or accept changes; a person with authority signs off on every position.
Review a draft agreement against your precedents and playbook
Feed a draft agreement, a few chosen precedents and your negotiating playbook to an AI that reviews it from several angles, has the findings challenged, and returns a memo plus proposed tracked changes for a lawyer to accept or reject. Sullivan & Cromwell built its Agreement Analyzer this way with OpenAI.
Act as a careful first-pass reviewer for a lawyer. Draft agreement: [paste]. Precedents to compare against: [paste 1-3 prior agreements or clauses]. Our playbook positions: [paste]. Client priorities: [list]. Review the draft in three separate passes: (1) deviations from our precedents and market practice, (2) risk allocation and how clauses interact, (3) ambiguities that could cause a dispute. Then challenge your own findings: for each, say whether it holds up, should be dismissed, or is unverified. Output a short memo grouped by confidence, quoting the exact clause and the precedent or playbook line behind each point, and a table of proposed redlines (current text, suggested text, reason). Do not state market practice without pointing to a source I gave you. This is not legal advice; a lawyer will decide.
Prep an anti-money-laundering alert review with every source linked
Have an agent pull together the evidence for a suspicious-activity alert, compare it to what is normal for that customer, and draft the case summary so the analyst only has to decide. FIS made its AML Investigations Agent, built with Anthropic, available to US banks; at design partner Amalgamated Bank it is projected to cut first-level alert review time by more than 90% and SAR narrative prep by more than 80%, and no case can be closed without an analyst confirming it.
You are helping me, a compliance analyst, prepare a review of one transaction-monitoring alert. Customer profile: [customer type, business or occupation, how long they've banked with us, normal monthly activity]. Alert: [alert type, date, rule that fired]. Transactions: [paste the flagged transactions plus the last [3/6] months for comparison]. Step 1: describe what is normal for this customer and how the flagged activity differs, e.g. a seasonal cash spike versus deposits kept just under reporting thresholds. Step 2: list every fact you rely on with the exact transaction or document it came from. Step 3: list the red flags and the innocent explanations side by side. Step 4: suggest one of: close with rationale, escalate for deeper review, or consider a SAR, and draft a short neutral narrative. Use only the data I gave you, mark anything missing as 'needs info', and remember the final decision is mine.
Process product complaints and flag the ones regulators need to hear about
For companies that make regulated products (medical devices, pharma, food, consumer goods), an agent reads each incoming complaint, pulls out what happened, flags any harm or hazardous situation, and prepares the complaint record so the quality team only reviews and signs off. Honeywell made TrackWise Digital's AI Processing Agent generally available on Oct 21, 2026, after five months in production at a medical device company where it processes thousands of complaints a day.
You are a complaint intake assistant for our quality team. Our products: [list]. Our complaint categories and reporting rules: [paste your SOP: categories, what counts as an adverse event or serious injury, reporting deadlines]. For each complaint below: Step 1: summarize what happened in 2 sentences (product, lot/serial if given, date, what the customer reported). Step 2: say whether there was patient or user harm, a hazardous situation, or a malfunction that could cause harm, quoting the exact words that support it. Step 3: suggest a category and whether it may need regulatory reporting under our rules, with your reason. Step 4: list missing information we need to ask the customer for. Mark anything involving injury or possible reporting as 'needs human review today'. Don't make the final reporting decision, and never fill in facts that aren't in the complaint. Complaints: [paste]
Pull every promise out of a client contract and turn it into an owned, dated checklist so nothing gets missed
Contracts and statements of work bury commitments such as reporting dates, staffing levels, service levels and penalties across dozens of pages, and missing one can cost money and reputation. AI agents can read each contract, list every obligation with the page it came from, assign an owner and due date, and flag clauses that conflict. Microsoft reported on Oct 21, 2026 that Birlasoft built a system of six specialist agents, coordinated by a planning agent, that extracts and validates obligations from more than 1,000 statements of work a year, finding over 10 obligations per project on average and saving about 4,000 person-hours a year, with people reviewing the results because the obligations carry financial and legal weight.
Here is a client contract or statement of work: [paste or attach]. Our team and roles: [paste names and roles]. 1) List every obligation we have, one per row: what we must do, deadline or frequency, who on our side should own it, the penalty or consequence if missed, and the section and page it comes from with a short quote. 2) List the client's obligations to us the same way (payments, approvals, access, data). 3) Flag clauses that link to each other across the document, such as a term on one page that changes a deadline later. 4) Flag anything vague, unusual or in conflict, and suggest a question to ask the client. 5) Give me a 30-60-90 day checklist of what's due first. Quote the contract exactly; don't guess at anything not in the text. This isn't legal advice, so note anything a lawyer should review.
More in AgentPlaybook Pro
- Contract risk review (first pass)Spot risky clauses before you talk to a lawyer.
- Benchmark a contract against your past dealsCompare a new contract's key terms with what you've agreed before, with the same counterparty and across all your contracts, to spot what's unusual, like Portfolio Benchmarks in Ironclad Agent.
- Run a first-pass ISO gap analysis of your company documents, with every finding linked to the page it came fromBefore an ISO 9001, ISO 27001 or similar audit, small companies often pay a consultant to compare their policies and procedures with each clause of the standard and list what's missing. An AI assistant can do a first pass: go clause by clause through your documents, mark each requirement as covered, partly covered or missing, and quote the exact document and section as evidence, so you know where you stand and what to fix before paying for expert review. Microsoft's Oct 23, 2026 customer story on Everstar says its Claude-powered Gordian platform delivered an ISO gap analysis that would have cost $20,000 and taken weeks for $1,050, with every claim linked to a citation in the source documents and qualified engineers reviewing the output.
32 copy-paste agent prompts (PDF) plus 5 new prompts every week.
Get it free