A

Align Research

The AI is the product. Align Research sells one job: its AI agents take a plain language question, search, read the retrieved opinions and return the relevant cases with the key passages marked. Without the agents there is nothing to buy, since the opinions themselves are available elsewhere and the product adds no editorial content of its own.

Grounding is built into the design, and accuracy is not measured. The product returns only existing court opinions with complete citations the lawyer can check, marks the passages it relied on, and composes no text, so it cannot put an invented case or quotation in front of a reader. What it can do is miss a case or mark the wrong passage, and the vendor's own terms say outputs may contain errors. No recall, precision or relevance measurement is published, so a lawyer cannot tell how often a research job leaves out a case that should have been found, which is the failure that matters for a retrieval tool.

A clear position that the lawyer decides, with real review surfaces. The vendor says the product leaves legal analysis and judgment with the attorney and will not summarize a holding or tell the lawyer what to argue, and its paid terms make the customer responsible for reviewing outputs before relying on them, including in providing legal services. Each returned case carries a relevance tier set by the model that the lawyer can change, with the relied-on passages marked.

What is not published is how the agents decide, during a job that runs for hours, when they have searched enough or which cases to drop before the lawyer sees the set.

No production evidence for this product is published. Align Research became generally available on August 3, 2026, and nothing published names a firm or legal department using it or reports a result. The named law firm testimonials on the Align site, from Whiteford, Taylor & Preston and Babin Bessner Spry, concern Align's trial binder product.

Strong contractual commitments on the paid tier, without coverage of privilege or the model provider. The Paid Tier Terms of Service, effective July 9, 2026, limit the vendor's use of questions and results to performing the service. They bar it from training or improving any model on that content, and a breach of that promise sits outside the liability cap. Support staff may view content only transiently to resolve a request, customers can delete items at any time with deletion within 30 days, and the data processing addendum commits to logical separation between customers.

The free tier works the other way. Its terms and the privacy policy let the vendor keep free tier questions indefinitely and use them to train its models, so a lawyer trying the product on a live matter should know which tier they are on. Nothing names the model provider or says what it may retain, and privilege and work product are not addressed directly.

A real position on tool versus judgment is published, short of the full picture. The vendor positions the product as retrieval only: it will not summarize a holding, draft a brief or tell the lawyer what to argue, and the judgment stays with the attorney. The paid terms put review of outputs on the customer before any reliance, including in providing legal services, and define authorized users as the customer's partners, associates, paralegals and staff.

What is missing is a statement on who may use the free tier, which is open to anyone resident where the service is offered, and any treatment of supervision when a paralegal runs a job.

No governance position for the AI is published. The retrieval only design is a deliberate safety choice that the vendor explains at length, but nothing names who is accountable for the agents' output, what is tested before a release, or how coverage and relevance judgments are checked for gaps across courts, jurisdictions or areas of law. The company's security page describes a compliance committee for security and privacy, not for the AI.

Most of the ground is covered in the paid agreements, short of a published subprocessor list. The data processing addendum limits processing to performing the service, lists security measures including access controls, encryption in transit and at rest, logical separation between customers and backup and recovery, commits to breach notice without undue delay, and deletes customer data within 30 days of a deletion request or termination, with backups cycling out in the ordinary course.

Third party audit reports such as SOC 2 are offered on request once a year. Subcontractors are bound by equivalent terms, but their list is available only on request.

A real position is published, with a notable carve-out and no stand behind the output. The paid terms cap each party's liability at the greater of 12 months of fees or $100, but leave breaches of confidentiality, the no-training promise and the indemnities outside the cap, so misuse of a firm's research questions is not limited to what the firm has paid. The vendor defends the customer against claims that the service infringes intellectual property.

For the output itself it gives no accuracy warranty and says results may contain errors the customer must review, and its 99.5 percent availability commitment carries no service credits.

No integration into practice systems is published for this product. A lawyer signs in, optionally with Google, submits a question, and receives results by email and in the application. No document management, Word, practice management or research platform connection is named. Align's trial binder product works with single sign-on and device management and exports to a document system, but those connections belong to the binders.

Cloud delivery is stated and location is not. The privacy policy says Align Research is a software as a service product processed in the vendor's cloud, unlike the binder product, which firms host themselves. The data processing addendum commits to logical separation between customers, but no hosting provider, region or processing location is named for the research product, and the addendum refers only to lawful transfer mechanisms for cross-border processing.

Certification is stated, with the scope for this product left implicit. Align states a SOC 2 Type II audit covering security, availability, processing integrity and confidentiality, annual independent penetration testing, and quarterly compliance reviews with an independent security expert, and the Align Research launch release repeats that the company is SOC 2 Type II certified. The security pages describe the binder products, and nothing states that the research product is inside the audit's scope; the data processing addendum offers third party audit reports on request where available.

The vendor refers to its AI agents and its own models without saying what runs underneath. No model or model provider is named, where inference runs is not stated, and nothing commits to notice of change. The privacy policy refers to the vendor's AI models being trained on free tier content, and none of those models is identified.

A buyer can learn the full cost without a sales process. Each research job costs $100, paid through prepaid credits at one credit per job, with no seat minimum, no annual contract and nothing to implement, and a separate free tier gives three jobs a month. The vendor frames the price as a per task cost a firm can bill back to the matter as a line item. Credits are non-refundable and a job cannot be canceled once it starts.

Who it is for and what it does not do are stated plainly; how far the case coverage reaches is not. The product returns published and unpublished US federal and state court decisions, is aimed at litigators at small and midsize firms and at general counsel and legal operations teams, and states that it returns cases only, with no summaries, drafting or analysis. It does not say which courts or date ranges the underlying case collection covers, which a litigator needs to know before treating a search as complete.