AI Agent Due Diligence: What Buyers, Investors and Enterprises Should Review
Evaluate the technical product, IP chain of title, foundation-model dependencies, data rights, APIs, open-source components, security, contracts and scalability before buying, licensing, investing in or deploying an AI-agent business.
What should AI-agent due diligence cover?
Primary-source basis: WIPO's venture guidance describes IP due diligence as a structured audit of patents, trademarks, copyrights, trade secrets and related intangible assets used by investors, buyers and partners to identify strengths and risks. WIPO IP due diligence guide.
10-Layer AI Agent Due Diligence Stack
| Layer | What to verify | Typical red flag |
|---|---|---|
| 1. Product | Architecture, deployment, reproducibility, supportability | Demo depends on manual intervention not disclosed to buyer |
| 2. Source code | Ownership, repositories, assignments, contributor history | Founder or contractor code never assigned to company |
| 3. Patents | Applications, ownership, inventorship, filing scope, prosecution | Core technical claim not owned or not filed in key market |
| 4. Models | Provider terms, weights, API use, downstream restrictions | Commercial product depends on rights that cannot be transferred |
| 5. Data | Training, retrieval, customer, licensed and generated data rights | Company has access but not commercial reuse rights |
| 6. APIs and tools | Third-party terms, rate limits, resale rights, continuity | Critical feature fails if one external API is terminated |
| 7. Open source | Licenses, notices, copyleft exposure, SBOM | Distribution model conflicts with license obligations |
| 8. Security | Identity, authorization, tool permissions, logging, incident controls | Agent can execute privileged actions without traceable authority |
| 9. Contracts | Customer rights, warranties, SLAs, assignment/change-control clauses | Key contracts terminate or require consent on acquisition |
| 10. Commercial resilience | Margins, customers, churn, platform dependence, scalability | Revenue depends on one customer or one external platform |
Start with ownership and chain of title
The first diligence question is whether the company owns the assets it says it owns. For an AI-agent business, that can require more than reviewing a patent schedule. The buyer should trace source code, inventions, prompts, orchestration logic, workflow design, confidential know-how, brand assets and technical documentation back to founders, employees and contractors.
Written assignment evidence matters. A company may control a code repository operationally while still lacking clean legal title if a founder developed core code before incorporation or a contractor retained ownership under the original services agreement. Acquirers should therefore review employment and consulting agreements, invention-assignment documents, IP schedules and material contributor history.
Patent diligence should connect claims to the actual product
A patent portfolio is more valuable when the buyer can map important claims to features that create customer value or competitive differentiation. Diligence should identify granted patents, pending applications, priority relationships, inventorship, assignment record, jurisdiction coverage, office-action status and known prior-art risks.
For AI-agent products, the buyer should distinguish broad marketing descriptions from the technical mechanism actually protected. A filing directed to generic machine learning may provide limited strategic value if the commercially important feature is a specific orchestration, security, memory or tool-execution architecture not covered by the claims.
Foundation-model dependency can be a transaction-level risk
Many agent businesses own the application layer but depend on hosted foundation models. Diligence should identify each provider, contract, API dependency, use restriction, rate limit, data policy and termination right. The key question is whether the product can continue operating if the provider changes pricing, terms or access.
A buyer should also distinguish contractual allocation of inputs and outputs from ownership of the underlying model. Provider agreements can grant commercial use of outputs while retaining provider ownership of the service and model technology.
Data diligence should separate access from rights
Training data, retrieval corpora, customer data and licensed databases can each carry different rights. The company should be able to explain where the data came from, what agreement governs it, whether commercial reuse is permitted, whether customer data can be sent to third-party providers, whether retention occurs, and whether a buyer can continue the same use after a transaction.
For retrieval-based agents, the buyer should also review whether the company can expose source content to end users and whether database or contractual restrictions limit downstream display.
APIs, tools and MCP integrations should be mapped as dependencies
An agent may depend on search APIs, messaging systems, databases, payment tools, developer platforms, browser automation, MCP servers or other services. Due diligence should identify which of those dependencies are essential, whether they can be replaced, whether the company has rights to resell or expose them through a customer-facing product, and whether a change of control affects the relationship.
Open-source diligence should include an SBOM and license review
Open-source software can accelerate product development, but a buyer should know which components are present and which obligations attach to distribution. A software bill of materials can help identify dependencies, versions and licenses. Particular attention should be paid to copyleft obligations, attribution requirements, modified components and embedded code that may be distributed to customers.
WIPO's due-diligence guidance specifically recommends maintaining software bill-of-material information and a structured data room so buyers can trace relevant IP and contractual evidence efficiently.
Security diligence is different for agents that can act
An agent that can send email, change records, deploy code, call APIs or operate enterprise tools creates a different risk profile from passive software. The buyer should understand identity, authentication, authorization, least privilege, approval gates, tool permissions, logs, rollback and incident response.
Security review should identify whether agent actions can be attributed to a human or service principal, whether privileged actions require approval, whether external content can manipulate the agent, and whether memory or context can be poisoned or exfiltrated.
Enterprise contracts can affect acquisition value
Customer contracts may contain assignment restrictions, change-of-control provisions, service-level obligations, pricing protections, data-processing commitments, indemnities or termination rights. A buyer should know whether the transaction itself triggers consent or creates new obligations.
Marketplace contracts also matter. If revenue is generated through Google Cloud, Microsoft, AWS, Salesforce, ServiceNow or another platform, diligence should examine seller eligibility, billing structure, current marketplace terms and whether the seller account or listing can continue after the transaction.
Scalability should be tested against the real cost structure
Revenue growth does not automatically mean improving economics. Agent products can incur variable costs from foundation-model calls, specialist data, tools, compute, observability and customer support. Due diligence should compare customer revenue with the costs created by heavy usage and with any minimum commitments owed to providers.
Customer concentration can create valuation risk
If one or two customers account for a large share of revenue, the buyer should examine renewal dates, termination rights, custom features, pricing discounts and support obligations. A high-value customer can improve traction while simultaneously creating concentration risk if the relationship is difficult to transfer or economically unattractive.
Acquisition diligence should test transferability
For an acquisition, the buyer should ask whether the important assets and relationships can move with the business. Source-code ownership, patents, data licenses, model agreements, APIs, marketplace accounts, domain names, customer contracts and key employee knowledge may each have separate transfer mechanics.
AI Agent Red-Flag Matrix
| Red flag | Why it matters | Possible transaction effect |
|---|---|---|
| Unassigned founder or contractor IP | Company may not own the core asset | Closing condition or valuation reduction |
| Non-transferable model/data license | Product may lose a critical input after acquisition | Restructuring or replacement dependency |
| No SBOM | Open-source exposure cannot be quickly assessed | Extended diligence and remediation |
| Critical single API dependency | Supplier can impair the product | Continuity discount or contingency planning |
| Broad customer indemnity | Risk can exceed contract value | Escrow, cap negotiation or valuation adjustment |
| Weak action logging | Enterprise incident accountability is limited | Security remediation before deployment |
| Marketplace-only revenue concentration | Platform changes can affect distribution | Channel diversification requirement |
Buyer Evidence Request Checklist
- Corporate IP ownership schedule.
- Founder, employee and contractor invention assignments.
- Source-code repository ownership and contributor history.
- Patent and trademark schedules with assignment records.
- Model-provider contracts and current commercial terms.
- Training, retrieval and licensed-data agreements.
- API, tool, MCP and platform dependency list.
- Software bill of materials and open-source notices.
- Security architecture, permission model and incident history.
- Material customer contracts and change-of-control terms.
- Marketplace seller agreements and listing information.
- Revenue concentration, gross margin and variable-cost data.
- Support and service-level obligations.
- Known disputes, infringement allegations or takedown issues.
Acquisition Data Room Index
A practical data room can be organized into corporate ownership, patents and trademarks, source code, employment and contractor agreements, open-source records, model and data licenses, customer and partner contracts, security documentation, marketplace arrangements, litigation or claims, product architecture, financial metrics and commercial dependency schedules. The purpose is not volume. It is traceability from each important business asset to the evidence showing ownership, permission, transferability and risk.
Frequently asked questions
What is the biggest AI-agent due diligence risk?
There is no single universal risk, but chain-of-title gaps and critical third-party dependencies are especially important because they can undermine whether the buyer actually receives a commercially usable product.
Does a patent portfolio prove that an AI-agent company owns the product?
No. Patents cover claimed inventions. The company may also depend on source code, trade secrets, data, APIs, model-provider rights, trademarks and customer contracts that need separate ownership and transfer analysis.
Should buyers review open-source software?
Yes. Buyers commonly review the software bill of materials, licenses, notices, modifications and distribution obligations to identify restrictions or remediation needs.
Why are model-provider agreements important in due diligence?
The product may depend on a provider's model, API or commercial terms. A buyer should understand permitted use, data treatment, termination rights, transferability and whether the dependency can be replaced.
What is different about acquisition due diligence?
Acquisition diligence should test not only present-day compliance but also whether the important rights, contracts, accounts and dependencies can transfer or continue after a change of control.
AI Agent Due Diligence: What Buyers, Investors and Enterprises Should Review — patent and commercialization context
Explore how the commercial decision connects with ownership, licensing, patent rights, and the cross-border protection of AI-agent technology.
Continue through the AI-agent commercialization cluster
Due diligence should connect technical reality to legal ownership and commercial transferability.
International Patents connects patent strategy, technology commercialization, transaction diligence and cross-border IP planning for emerging technologies.
Discuss AI-agent acquisition, investment or enterprise due diligence.
Email: rd (at) patentbusinesslawyer (dot) com