DataGrail

AI Centrality

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.

A named, genuinely integrated agent sitting on a substantial conventional platform. Vera is marketed as the organising idea of the product, with the company positioning itself as an Agentic Data Privacy Platform, and its functions are specific rather than gestural: proactively detecting and investigating new cookies and proposing consent rules, autofilling PIAs, DPIAs, AI risk assessments and transfer impact assessments from platform data, flagging risks across a large application catalogue with action plans, and answering questions on demand. That is a real layer and it is described in current first-party material. It is not the mechanism being bought. Strip Vera out and the platform still does its job: the Live Data Map, Request Manager, Consent Management, assessments and Risk Register are the deterministic products the company sold before the agent existed, and the privacy policy describes the AI features as customer-initiated features within the Services rather than as the Services. The company's own framing confirms the placement, describing Vera as making existing privacy jobs faster rather than as performing a task the platform could not otherwise do.

Citation Accuracy and Hallucination Disclosure

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.

Accuracy is asserted without measurement and grounding is described architecturally rather than technically. The grounding claim is real and is the product's central pitch: assessments are described as evidence-based because they draw on the Live Data Map's live system inventory rather than on a survey, and the contrast is drawn explicitly against a static data mapping exercise that is out of date the moment it is complete. Vera is said to employ global privacy context to produce accurate PIAs and DPIAs. None of that is a measurement. No accuracy figure, no test set, no evaluation, no error rate and no benchmark of any kind was located, and hallucination is not addressed anywhere on any surface read under any name, which is a notable gap on a product that drafts the assessments a regulator may later read. Nothing describes what happens when Vera's autofilled assessment is wrong, and no correction or review record is described beyond the general approval gate.

Autonomy and Oversight Model

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 approval gate is published as a design principle and appears in the product description rather than only in marketing. The clearest statement is operational: Vera will detect new cookies, investigate them and suggest rules, and apply them **when you give the word**, which places the acting step behind an explicit human decision. The home page names the principle twice more, as AI-powered action with human control and as secure, human-governed AI, and the privacy policy independently confirms it from the legal side by describing the AI-enabled features as customer-initiated. One published control is unusual enough to record: a six-stage prompt protection scheme, which addresses prompt injection, a real risk for an agent reading third-party web content and inbound requests. What is absent is the rest of the structure. No threshold is published at which Vera acts alone, nothing distinguishes which of its actions are reversible, no confidence signal is described, and the six stages are named as a count without being explained.

Operational and Outcome Evidence

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.

Named customers with attributed quotes and real figures, short of dated results. The logo wall is unusually substantial and identifiable, including Major League Soccer, HubSpot, FanDuel, Netgear, Reformation, Dexcom, Okta, GoFundMe, Life360, Vercel, Databricks and Zillow. More importantly the testimonials carry names, job titles and employers rather than initials, and several of those titles are the buyer this index cares about: a Vice President Legal at Poppulo, a Senior Privacy Analyst at nCino, and Associate General Counsel and Data Protection Officer titles quoted on the customers page. Figures are attached to named deployments rather than floating free: Major League Soccer unified privacy across 30 independent clubs, mapping more than 2,500 systems and automating more than 200 data subject requests, and a named customer reports consent policy review time down by over 75 per cent since switching. What holds this below the top band is dating. No result carries a date or a measurement period, and the customer case studies were not opened in this pass.

Privilege and Confidentiality Posture

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.

Substantive commitments across most limbs, none of them contractual, because the agreement that would carry them is not published. What is published is specific and unusually well aimed. Training: the home page states plainly that there is no training on your data, and the privacy policy adds the harder-edged half, that DataGrail does not permit AI service providers to use customer data to train their AI or machine learning models. Architecture: the platform is described as single tenant, and the security page records that customers provision cloud storage in their own environments with limited permissions granted to DataGrail, which keeps much customer content outside the vendor's estate entirely. Roles are stated properly, with DataGrail the processor of all Customer Data and the controller only of visitor and account data. Encryption is AES-256 at rest and TLS 1.2 in transit. Two limbs are missing. Privilege and work product are not addressed at all, which matters because the buyer is frequently in-house counsel. And the qualifier on DataGrail's own use is recorded rather than glossed: customer data is used to generate outputs **and to operate and improve the Services**, a phrase that does not name training but is broader than the marketing line suggests.

UPL and Professional Responsibility Posture

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.

A disclaimer exists and is expressly scoped away from the product. The terms of service carry a dedicated clause headed as a disclaimer that the offering is not a substitute for legal advice, stating that the content, products, resources and services made available **through the Site** are not intended to provide legal advice. The limitation is in the words: those same terms open by stating that they expressly do not govern the subscription services offered through the DataGrail Platform. So the advice disclaimer covers the marketing website and not the software that autofills data protection impact assessments and generates recommendations for a legal team. Nothing published addresses where the tooling stops and legal judgement begins in the Platform itself, nothing describes what a privacy lawyer must review before an AI-drafted assessment is relied upon, no professional obligation is named, and no jurisdiction limit is stated for the regulatory conclusions the product reaches. The gap is structural rather than an oversight of wording, since the document that would carry it is not published.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

Real published positions on AI conduct, with no framework, owner or testing behind them. Four specific commitments exist and are more than most records in this band show: no training on customer data, no permission for AI service providers to train on customer data, a six-stage prompt protection scheme, and human governance of agent action. The company also sells AI governance as a product, with a DataGrail for AI Governance solution and AI risk assessments in the assessment suite. That last point cuts both ways and is worth naming: a vendor whose product helps customers govern their AI publishes no governance framework for its own. There is no responsible AI page, no named owner accountable for Vera's behaviour, no description of evaluation or red-teaming before release, no published results, no model card, and nothing at all on bias, which has a concrete shape here since Vera's risk flags and assessment drafts shape what a privacy team investigates. The six stages are asserted as a number and never described.

AI Safety and Data Stewardship

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.

Substantive and specific across most of the set, with one limb vague and the primary surface stale. Access control is well covered: two-factor authentication with Okta, Google SSO and other providers, customer-provisioned storage with limited permissions granted to DataGrail, and AES-256 at rest with TLS 1.2 in transit. Resilience is quantified rather than claimed, with daily encrypted backups and a stated 24-hour recovery time objective, plus a public status page carrying a record of past incidents. Testing is committed to a cadence: penetration tests every six months with issues handled within a day, alongside a bug bounty programme. Service providers are named at length in the privacy policy across hosting, analytics, sales and payments, and the AI processing route is identified. Transfers use Standard Contractual Clauses or the EU-US Data Privacy Framework. Two things hold it at B. Retention is the weak limb, committing only to keeping data as long as necessary, with anonymised data retained as long as DataGrail itself determines is commercially necessary. And the security page carries a last-modified date of November 2022, four years before this check, so it predates the agentic product entirely and none of the single-tenant, no-training or prompt-protection claims appear on it.

AI Liability and Recourse

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.

No agreement governing the product is published, and the vendor says so itself in the first paragraph of the only agreement it does publish. The terms of service state that they govern access to and use of the datagrail.io website and that they **expressly do not govern the subscription services offered through the DataGrail Platform**, which are instead subject to a Master Services Agreement to be executed between DataGrail and its subscribing customers. That agreement is not published anywhere. Everything the site terms do contain therefore allocates loss for website use only: the as-is disclaimer, the exclusion of consequential, incidental, indirect, exemplary, punitive and special damages with no cap stated, and an indemnity running only from the user to DataGrail. Applying the convention that an agreement governing something other than the product is an absence, this is graded as though nothing were published. No indemnity, no cap, no warranty on output, no service level and no insurance position for the Platform could be located, so a legal team buying a tool that drafts its impact assessments cannot read the allocation of loss before entering a sales process.

Practice Systems Integration Depth

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.

The integration network is the product's engine and it is documented at the level an implementer works from. More than two thousand connections are published, each with its own page rather than appearing as a logo, with Salesforce, Okta, Shopify, Zendesk and Webflow surfaced in the navigation and the full catalogue browsable. What the integrations do is described rather than asserted: they feed the Live Data Map so that systems holding personal data are detected and catalogued automatically as they come online, and they carry data subject request fulfilment out to connected applications so access and deletion happen in the source systems. Coverage is stated across third-party SaaS, internal databases and warehouses, and home-grown systems reached through a pre-built API. A separate developer documentation site is published, and identity integration is specific, naming Okta and Google SSO for authentication. The underlying mapping technology is covered by an issued United States patent, which is a verifiable fact rather than a claim. For this product class the connected systems are the marketing, support and data stack rather than a document management system, and that is the correct surface to integrate with.

Deployment Model and Data Residency

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.

Both limbs are stated and neither is developed. Tenancy is addressed directly, and in the form buyers ask about: the platform is described as single tenant, which for an AI product handling personal data inventories is the answer a security reviewer wants. Residency is stated in the privacy policy rather than in marketing, recording that DataGrail is a Delaware corporation with offices in the United States and that it collects, processes, transfers and stores data within the United States, with European transfers handled under Standard Contractual Clauses or the EU-US Data Privacy Framework. The architecture note on the security page adds a real distinction between processing and storage, since customers provision cloud storage in their own environments with only limited permissions granted to DataGrail. What is missing is choice and detail. No region options are offered, so a European buyer is told where the data goes rather than given an alternative, no cloud region is named, and the single-tenant claim appears once, as three words on the home page, with no supporting description anywhere including on the security page.

Security Certifications and Trust Center

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.

Certifications are claimed by picture and by picture only. Both the trust centre and the security page carry a heading reading Certifications followed by an image file of logos, with no accompanying text anywhere: no framework is named in words, no auditor or certifying body is identified, no scope statement, no observation period, no report date and no route to obtain a report at any access tier, not even on request. A badge image alone is not an attestation, and this is the clearest instance of that in the corpus, because the vendor has not written a single sentence about what it holds. What is real and does lift this off the floor sits elsewhere on the same page and is a practice rather than a certificate: penetration tests performed every six months with issues handled within a day, a bug bounty programme with a published reporting route, and a public status page carrying a record of past incidents. The trust centre itself is a hub of links to the privacy policy, terms, security page and data rights portal rather than a document portal. The security page was last modified in November 2022.

Model Supply Chain Disclosure

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.

One real disclosure that names the route and not the models. The privacy policy carries a dedicated AI Enabled Service Providers section, which is more than most vendors publish, and it states that DataGrail uses AI services available through Amazon Web Services to process customer-submitted information solely on its behalf and in accordance with its instructions, and that it does not permit AI service providers to use customer data to train their AI or machine learning models. The constraint is genuinely useful and the processing route is identified. What it does not do is identify the supply chain: naming the cloud through which AI services are obtained says where inference is bought, not whose model performs it. No model is named, no version or family is given, no model provider behind the AWS layer is identified, no location is stated for inference beyond the general United States processing commitment, and no commitment to notify customers when the model or provider changes was located. Vera is a product name for the agent rather than a disclosure of what powers it.

Commercial Transparency

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 information is published at any level, including the unit of charge. There is no pricing page anywhere in the site inventory: the primary navigation carries Product, Solutions, Customers, Resources and Company, and the footer carries product, solutions, team, regulation and resource groupings with privacy and terms links, and no pricing entry appears in any of them. Every call to action on every page read resolves to a demo request or a contact form. Nothing states whether charging runs per seat, per integration, per data subject request, per domain or per entity, and no figure, band or range appears. Older solution pages describe a self-service Privacy Control Center purchasable as an end-to-end solution or à la carte, which describes a purchasing shape without attaching any number, unit or term to it, and no self-service purchase route was located from any current page. Under the standing rule that pricing evidence must lift the axis off the floor before a pricing row is owed, no VendorPricing row is written. Checked home, platform, trust centre, security, terms, privacy policy and the full navigation and footer on 4 September 2026.

Firm and Practice Coverage

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.

Segmentation is published along three axes and is specific on each. By team, the product is sold separately to Legal and to Security, each with its own solution page, which is the clearest buyer disclosure of any record built in this lane. By regulation, coverage is enumerated rather than gestured at, with individual pages for the EU GDPR, the California Consumer Privacy Act, the California Delete Act, the Colorado Privacy Act and the Virginia Consumer Data Protection Act, plus a guide covering United States privacy laws generally. By use case, the pages run to responsible data discovery, records of processing activities, do-not-sell and share, AI governance, retail and managed services. Team size is addressed explicitly rather than left to inference, with the platform pitched at a privacy team of one through to twenty-one. What is missing is the boundary. Nothing states which regulations or jurisdictions fall outside the product, the enumerated regimes are United States law plus the GDPR with no other national regimes named, and no statement describes where the product stops or which organisations it is not built for.