Relyance AI maps where personal and sensitive data goes inside an organization and ties each flow to the obligation that governs it. Instead of questionnaires and manually kept inventories, it discovers data movement from several points at once, reading source code, watching runtime behavior, scanning contracts and connecting to systems directly. Its Data Journeys technology keeps the result as a live graph from code through cloud infrastructure to AI models, and its Data Exposure Graph joins sensitive data, the identities that hold permissions over it, and the behavior of AI agents.
Three product layers sit on top. Privacy automation builds records of processing activities and data maps from live flows and runs consent, assessments and data subject requests. AI governance inventories agents, models and MCP servers and can block an agent with too much reach in the CI/CD pipeline. Data security classifies sensitive information and watches it in motion. Setup is agentless. Customers include Notion, Logitech and Zscaler.
Relyance Inc. is based in San Mateo, California, and was founded by a data privacy lawyer and a machine learning engineer.
Side by side on the same 15 capability axes and 12 legal signals: where the two part company, and what neither one publishes.
Vendors in the same categories as Relyance AI whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
Ordered by how closely the graded profiles match across the 15 capability axes, inside the same categories. See how each axis is graded in the methodology.
All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The models are the mechanism, and the company draws the distinction itself. Its about page says it is not a legacy tool retrofitted with AI but was built from first principles for a world where data behavior is not deterministic. It says the old approach of scan, classify and lock no longer holds, because scanners can tell what data is but not where it came from or what it is doing. What is sold is the inference: code analysis that reads how data moves through a repository, runtime monitoring, contract analysis that ties obligations to flows, classification of sensitive data, and analysis of how AI agents behave, joined into one live graph.
There is no content asset underneath. The material is the customer's own code, systems and contracts, and the graph exists only because the models build it. The work replaced is manual inventories and questionnaires.
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Grounding is documented in the architecture and never measured. Discovery is described concretely, through code analysis, runtime monitoring, contract scanning and system integrations, with every data flow mapped to its business purpose, legal obligation and risk. That gives a traceable chain from a finding back to the source that produced it, the design claim the product rests on. No measurement of whether the inferences are right is published: no accuracy, precision or false positive figure, no evaluation or benchmark, and nothing on how a classification or an obligation mapping is validated.
The gap has weight here because a missed flow reads as compliance where none exists, and a wrong obligation mapping produces a record of processing activities that is confidently incorrect. Hallucination in the features that draft assessments is not addressed either.
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
The possible actions are listed clearly, and the control around them is not. Published material says a request can be resolved in real time with one of six outcomes, ranging from allowed through redacted to escalated or blocked, with the decision computed at the moment it matters rather than from static allow lists. The CI/CD integration flags a risky change before merge, which puts a decision point inside an existing engineering review.
Nothing states what triggers escalation rather than a block, what confidence or severity threshold applies, who reviews an escalated request, or what a privacy team must approve before an automated enforcement takes effect. The runtime layer that does the blocking and redaction is marked PRIVATE BETA in the vendor's own navigation. What ships generally is posture management that flags and maps, and no published agreement places a review obligation anywhere.
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
The reference base is broad, and the people quoted are senior and named. About two dozen logos appear on the home page, from consumer technology such as Notion and Yelp to data infrastructure such as Grafana and security such as Zscaler. Three customers speak on the record with name, title and a linked video: Chris Bender, CISO at ClickUp, Jason James, CIO at Aptos, and Karthik Chakkarapani, SVP and CIO at Zuora. The Zuora quote is substantive, describing the difference between a snapshot and continuous visibility.
Two deployment figures are published: more than 1,600 systems and 100 flows mapped in under two hours without agents, and six surfaces unified in one graph. Neither figure carries a measurement basis or a date, no figure is attached to a named customer, and the case studies are video rather than written.
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Handling controls are published, and the questions that matter most to a legal buyer are not answered. The trust center rules out using production data in development or testing except to debug a customer issue, and provides for disposal of customer data on request. Whether customer content trains models is left open, and the published material does not separate one customer from another. Neither privilege nor work product is mentioned anywhere, though the platform generates and stores the privacy assessments a counsel would often expect to be privileged.
No model provider is identified. The operative agreement is a master services agreement available only on request, so none of this can be tested against a contract a prospect can read.
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
Nothing published addresses the advice line, and the marketing runs the other way. There is no statement that output is not legal advice, nothing on supervision or verification duties, and no statement on jurisdiction, and the terms of use cover the website alone. The privacy teams page says Relyance uniquely combines legal expertise with technical discovery to close the gap between documented policies and actual data practices.
Company material foregrounds that one cofounder and chief executive is a practicing data privacy attorney. The outputs are statutory instruments: records of processing activities and assessments of data protection, transfers and legitimate interest, each a document a regulator may read and a controller must stand behind. The platform generates them and markets legal expertise without a published statement on the professional boundary.
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
The product governs AI, and nothing published governs the vendor's own. The AI Security Hub, the AI governance solution, blog material on governance failures and a survey on consumer AI trust all concern what a customer should do about its models. None of the pages read, which include the about page, the solution pages, the terms and the trust center, addresses the models Relyance itself runs. There is no responsible AI page or set of principles for the platform, no individual or function named as accountable for model behavior, no account of what is evaluated before a model or classifier change ships, and no ISO 42001 or equivalent certification.
The trust center's monitored controls are information security controls. Uneven output goes unaddressed on a system that classifies sensitive data and maps legal obligations across jurisdictions, languages and codebases of varying quality.
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Every element of stewardship is addressed, in a trust center with roughly fifty named controls monitored continuously by Secureframe. The controls cover change management, availability, confidentiality, vulnerability management, incident response and access, among others. Access to customer data runs on unique identifiers, passwords backed by a second factor, scheduled access reviews and removal on termination. Encryption is stated at rest and in transit.
Incident practice is an incident response plan with documented tracking of incidents. A network and application penetration test by a third party runs at least annually, with critical and high findings tracked to resolution, and vendor SOC 2 reports are collected each year. Retention follows what each customer agreed rather than a published period, and the subprocessor list sits behind a request.
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
The agreement exists and can be requested, and it is not published, so the allocation itself cannot be read. The terms of use on the website govern the website only, and say that a signed agreement covering the Relyance Platform supersedes them. Those website terms disclaim all warranties, exclude liability for damages of any kind, run indemnity only from the user to Relyance, and send disputes to arbitration under California law, none of which describes what a platform customer receives.
The master services agreement and the data processing addendum are both listed in the trust center behind a request flow. Cybersecurity insurance is published as a continuously monitored control. No cap, indemnity, warranty on output or service level can be seen before contacting the company.
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
Integration is the core mechanism, and it points at engineering systems rather than legal ones. Discovery runs across code repositories, runtime environments, AI models and third party processors, with an integration framework reaching the wider data ecosystem and enforcement running inside CI/CD pipelines. Setup is stated to be agentless, with a first map in hours. Almost no specific system is named: no repository host, cloud provider, data warehouse, identity provider or consent platform appears on any page read, beyond consent signals captured from the software development kits of consent management platforms.
No connector list, API reference or developer documentation is published. No integration into systems a privacy or legal function itself runs is described, with no document management, matter management or GRC platform named.
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
Where the platform runs, and how customers are separated within it, is not published. No cloud provider is named, no hosting region or country is stated for customer data, no residency option is offered, nothing separates processing location from storage location, and no tenancy model is described. Separation of development, staging and production is an internal environment boundary rather than a tenancy statement. The only geographic material concerns transfers of personal data under the privacy statement, not platform hosting.
The platform reads a customer's source code and runtime traffic and is sold to European privacy leaders, and it states nothing about residency.
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
A public trust center carries real substance, and the evidence behind it sits behind a request. Hosted by Secureframe and continuously monitored, it lists SOC 2 Type 2, ISO 27001 and Data Privacy Framework participation under compliance. Ten documents are named, among them the SOC 2 Type 2 report, the ISO 27001 certificate and a penetration test executive report, and roughly fifty controls are published with descriptions rather than as a badge wall.
The privacy statement and the responsible disclosure policy are freely available. Every report and certificate sits behind a request control, and the portal does not say whether access comes instantly on an email, after a click through agreement, or after review. No auditor or certification body is named, and no report period, issue date or scope is published. No request was submitted.
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
A proprietary engine is named and nothing underneath it is. The company brands its system Lyo, describing it as the world's first data defense engineer native to AI, and repeats that the platform was built on AI from the ground up. No model, version or provider is identified, and nothing says whether any component is built in house or reached through a third party. Nothing describes where inference runs, what any provider may retain, or whether customers would hear of a change in the model set.
The subprocessor list that would carry the answer sits behind a request. The product's own job is to inventory every model, agent and retrieval system in a customer's environment and say which ones touch sensitive data, and the vendor discloses none of its own.
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
No pricing appears at any level. There is no pricing page in the navigation or the footer, no tier or plan is named, no unit of charge is identified, and no structure is described, whether per user, per system, per data source, per scan or by subscription. Nothing says whether implementation or onboarding is charged separately, and no trial or free tier is offered. The only commercial routes are a demo request and an offer of a free exposure map, both leading to the same booking form.
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Regulatory coverage is named precisely, and the buyer is described inconsistently. The platform is said to support GDPR, CCPA and the EU AI Act among other frameworks, with the consent layer realigning purposes as they change, and the blog tracks individual state statutes such as Maryland's. Buyer documentation wavers. A dedicated page addresses privacy leaders in their own terms, and the blog has a category for privacy and legal professionals.
Yet the home page is headed as an AI data security and governance platform trusted by security and governance teams, and all three customer voices are security and IT leaders. The site does not settle whether the privacy function, the security function or both buy the platform. No organization size band is stated, no industry pages exist, and nothing describes where coverage stops.
The public pages on file for Relyance AI, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.
Fabricated Citation Record
Read Sep 2, 2026
Third Party Request and Subpoena Notice
Read Sep 2, 2026
Relyance AI's 21 Sept release tightened Strict InHost mode so sampled record values and error text stay out of scan results. InHost runs the Relyance scanner in a cloud account the customer owns and copies only the results back to Relyance. The same release adds sanitized supporting context to data classification findings, which previously showed no context at all.
A signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.
Can material a lawyer puts into this product be used to train a model?
A published agreement or policy exists and none of it addresses the question either way, or the document that would answer it could not be read and the summary names the retrieval limit. The summary states which shape the silence takes: an improvement right granted that never names training, or no improvement right granted at all.
Training on customer content is not addressed in either direction. No clause or statement names training, model training, machine learning or model improvement in relation to customer data. The nearest provision is in the privacy statement, which permits use of personal data to operate the platform, provide related services and develop and improve the Relyance Platform and other services. That is an improvement right that does not name training.
The product itself sells the answer to this question, with a capability framed as proving that customer data was isolated and never used for AI training, logged as verifiable evidence. The vendor offers customers that assurance about their own vendors, and as of 2 September 2026 publishes no equivalent statement about itself.
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is acknowledged in public materials with no stated period.
Retention is acknowledged in two places and quantified in neither. The privacy statement keeps personal data only as long as needed for its stated purposes or as the law requires, then destroys it under retention policies. The trust center publishes a data retention and disposal policy, a control retaining customer data to agreed customer requirements, and a control removing data no longer needed on customer request.
A practice exists and is monitored, and the period is set per customer in the agreement rather than published. Nothing states how long discovered flow data, generated assessments or scan results persist, and no retention setting is offered in the product.
Does retrieval respect the firm’s ethical walls, or can the model read across them?
No located public material addresses walls or matter level segregation.
Segregation between customers is not described, and no tenancy model is stated. Two adjacent controls are published, and both concern internal boundaries. Development, staging and production environments are segregated, and access to customer data is limited to personnel who need it on a least privilege basis, with unique identifiers and scheduled access reviews. Those separate Relyance staff and systems from customer data, not one customer from another.
The question carries weight here, because the platform holds a complete map of a customer's code, data stores, identities and AI systems, close to a blueprint of the organization's attack surface. The terms, privacy statement and trust center were read in full on 2 September 2026.
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Published terms or policy address disclosure to authorities or in response to legal process, and no commitment or reservation regarding customer notice is located anywhere. The vendor has told the customer that data can leave and has said nothing about whether the customer hears of it.
Compelled disclosure is addressed, and notice is never reached. The privacy statement says Relyance may share personal data, as quoted, to respond to a lawful order such as a subpoena or to comply with other applicable law. It may also disclose personal data to government agencies to protect its legal rights or to deter fraudulent, unethical or illegal activity, a ground resting on its own judgment rather than on compulsion.
Nothing commits to telling the customer that a demand has arrived, to seeking confidential treatment or to letting the customer intervene, and no transparency report was located. The master services agreement and data processing addendum, available on request, were not requested and may address it.
Where does the law in this product come from, and does the vendor have the right to use it?
No located public material identifies the corpus behind the product’s answers.
No collection of its own is claimed on any page read on 2 September 2026. The material the platform works on belongs to the customer, source code, runtime traffic and contracts among it, all discovered in place rather than gathered into a collection Relyance holds. Provenance and licensing questions therefore sit with the customer. One dependency goes undescribed. Mapping flows to legal obligations and realigning them as regulations change requires a body of regulatory content, and no source, publisher, update cadence or license basis is identified for it.
Does the product tell you when the authority it just cited has been overruled?
No located public material addresses whether authority is checked for subsequent history.
Whether the law behind an obligation mapping is current, and how a user would know, is not addressed. The product retrieves no primary law for a reader, so a citator in the usual form has nothing to work on, but the analogous question is live. The platform says it turns regulatory requirements into operational guardrails, ties every flow to its legal, contractual and policy requirements automatically, and updates obligations as regulations change.
As of 2 September 2026, nothing states how quickly a change is reflected, which jurisdictions are monitored, who maintains the regulatory content, or whether a user sees the version or effective date of the rule an obligation rests on.
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
As of 2 September 2026, nothing describes what the platform does when it cannot reach an answer, and no confidence indicator is shown for any classification, flow mapping or obligation. Compound risk is scored across the graph, but that measures the customer's exposure rather than the system's confidence in its own finding. Escalation is one of the enforcement outcomes, and nothing states what sends a request down that path.
A published third party review notes that identified risks are not ranked; it is not the vendor's own material and is not relied on.
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
No court order, opinion or disciplinary record naming the product was located. The AI Hallucination Cases database maintained by Damien Charlotin, and reporting drawing on it, were searched on 2 September 2026 for Relyance AI and Relyance Inc. The result describes the public record rather than the product. The platform produces data maps, records and enforcement decisions rather than legal citations, so its characteristic failure is a missed data flow or a wrong obligation mapping.
Such a failure would surface as a regulatory finding, an enforcement action against the customer or a breach, none of which any tracker indexes the way the database indexes court sanctions.
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No located public material engages with bar or ethics guidance.
Professional conduct guidance was not engaged on any page read on 2 September 2026. No bar association material or ABA Formal Opinion 512 is referenced, and nothing addresses the competence, supervision or candor duties of counsel relying on an assessment or a record of processing activities the machine generated. What the vendor engages at length is legislation, the frameworks the product helps a customer comply with, and the company takes part in privacy profession events.
Naming the law a product serves differs from naming the rules that govern the professional using it, and only the first is published.
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The product does not touch a fee between a lawyer and a client. It operates before an engagement exists, or it is bought by a team that bills no client for the work. Savings claims aimed at the buyer’s own cost are recorded in the summary and do not make the row a savings claim, because no client bill is in the loop.
Savings are claimed in headcount and time, and the bill is never addressed. The published pitch is that a customer can meet every privacy obligation without growing the privacy team, with continuous discovery replacing manual inventories and questionnaires. The buyer is an in house privacy or security function rather than a firm billing a client, so the savings accrue to the customer's own headcount rather than to an invoice.
The question reaches outside counsel indirectly. Assessments and records the platform generates are work privacy counsel would otherwise scope to a law firm, and nothing published addresses how that shift shows up in a firm's engagement, nor is any record of AI assisted work per matter described.
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The material exists behind a sales conversation or an executed agreement.
The material exists and is listed, and none of it is published. The trust center names a subprocessor list, a data processing addendum and a master services agreement among its documents, each reachable through a request rather than a download. Two items can be forwarded today: the privacy statement, which is public, versioned at 2.3 and links its previous version, and the responsible disclosure policy. Certification under the EU-U.S., UK Extension and Swiss-U.S. Data Privacy Frameworks can be checked separately on the Department of Commerce register.
No model provider is named anywhere, so the requested subprocessor list is the only route to learning which AI providers touch customer content.
If a judge’s standing order requires an AI disclosure, can the product produce one?
No located public material addresses court disclosure or verification certification.
Nothing read on 2 September 2026 addresses disclosure of the platform's own AI use to a court, regulator or auditor. No exportable record is described of which model or classifier produced a determination, on what confidence, or who reviewed it. Evidence generation is a headline capability pointed the other way: complete audit trails, defensible evidence of consent changes for regulators and auditors, every data touch logged as verifiable evidence, and proof ready for audit on demand.
All of that evidences the customer's data practices to a third party. None of it is described as showing how the platform itself reached a conclusion, which is what a customer would need if a regulator questioned a record of processing activities the system generated.
Every grade and every signal on this index is drawn from public sources and dated. If a record is wrong, out of date, or missing an artifact the index did not locate, send the source and it will be reviewed and the record redated. Vendors are welcome to submit documentation. Nothing on this index is for sale, including a listing, a placement, or a grade.