Darrow

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.

The models are the product and the agreement says so in terms. Section 8.3 states that the AI features are mandatory and integrated by default, and that a customer who does not accept the AI service providers' terms has no remedy other than to stop using the service. The published architecture is four layers of model work: ingesting raw signals at scale, interpreting them through legal context, reasoning across the result to prioritise and orchestrate action autonomously, and surfacing it to the teams responsible. Published throughput of more than 5 million signals detected monthly is not a volume a non-model system reaches. Remove the models and there is no product. Checked 4 September 2026.

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.

Grounding is asserted without a mechanism a reader can inspect. The knowledge base is described in principle, connecting the laws that create obligations, the organisational weaknesses that emerge when they are not met, and the enforcement actions that follow, and the inputs are described by category as regulatory filings, incident reports, litigation patterns and market activity. Nothing describes how an output traces back to the signal that produced it: no citation, no source link, no retrieval method. Hallucination is addressed, but only as a disclaimer: section 8.3 records that generative AI involves uncertainties as to quality and accuracy and that the company has no control over output, and the trial terms disclaim liability for any undetected, misdetected or mischaracterised events, signals, risks or cases. No accuracy figure is published, no test set is described, and section 12.1 expressly disclaims any warranty of accuracy. The Research and Domain Coverage pages were not opened in this pass and are named here so the limit is visible.

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.

Autonomy is claimed in plain terms and oversight is asserted without a mechanism. The published description of the agentic layer says advanced AI systems reason across the intelligence, identifying emerging opportunities, prioritising signals and orchestrating action autonomously, and the product is framed as an always-on legal red team. Against that, the only oversight statements are contractual disclaimers: section 8.3 provides that the customer is responsible for evaluating the results of its use of the AI features and that the service is not a substitute for legal advice, legal research and independent review. Nothing published describes what the system does unattended against what a person approves, no threshold is stated at which it escalates or stops, no review surface inside the portal is described, and nothing addresses what happens when a signal is wrong. Searched the home page and the terms of use on 4 September 2026; the platform, law firms and FAQ pages were not opened.

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.

Substantial deployment evidence, with the names and the numbers on opposite sides of the page. Roughly thirty customer organisations are shown by name, and they are serious ones in this market: Baron and Budd, Milberg, Girard Sharp, Bernstein Litowitz Berger and Grossmann, Zuckerman Spaeder, Singleton Schreiber, Kahn Swick and Foti, Lynch Carpenter, Bailey and Glasser, Nichols Kaster, Loevy and Loevy, Wolf Haldenstein, Stephan Zouras, Schneider Wallace, Freed Kanner, Kozyak Tropin and Throckmorton and Consovoy McCarthy, alongside Microsoft and Armis on the corporate side. Figures are published at corporate level: more than 22 billion dollars in legal exposure surfaced, over 80 organisations, more than 5 million signals detected monthly, over 22,000 legal professionals and more than 10,000 active matters. What is missing is the join between them. No figure is attached to any named firm, no case study is published, nothing is dated, and there is no testimonial from a named individual anywhere on the surfaces read.

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.

Nothing published addresses how client confidences are handled, and the terms that do exist run the other way. The confidentiality regime at section 11 defines Confidential Information as non-public business, product, technology and marketing information, and does not bring Customer Data within that definition, so the mutual confidentiality obligations do not by their terms reach what a customer puts into the service. Section 5.1 grants the company and its sub-processors an irrevocable, perpetual, sub-licensable licence over Customer Data that expressly extends to training foundation models. Section 11.3 reserves a broad right to access, read, preserve and disclose any information obtained in connection with the service. Nothing addresses privilege or work product, nothing addresses segregation between customers or matters, and no retention period exists. What does exist is recorded for completeness: section 7.3 commits to reasonable industry-standard technical and organisational security measures to protect Customer Data, which is a security promise rather than a confidentiality one. Section 5.2 separately prohibits the customer from submitting protected health information.

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 inside the terms and the product surfaces do not address the question. Section 8.3 states that the customer acknowledges the service is not a substitute for legal advice, legal research and independent review, and the trial terms go further, adding that it is not a substitute for audit, assessment or any medical, legal or regulatory conclusion and that the customer is solely responsible for obtaining necessary legal, medical, scientific or regulatory advice and for its own due diligence. Both sit inside AI-features and warranty clauses rather than in a dedicated position. Nothing on the home page or the segment pages addresses the question at all. No competence or supervision language appears, no rule of professional conduct is named, and jurisdiction is handled by disclaimer rather than by limit, with section 12.1 disclaiming any guarantee that output is suitable for the customer's jurisdiction. The product recommends which matters a firm should pursue, which makes the absence of a supervision position more material than the disclaimer acknowledges.

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.

No governance position is published for a system that decides which potential claims a firm sees. There is no responsible AI statement, no governance framework, no named owner accountable for model behaviour, no description of pre-release evaluation, no published testing results, and nothing on bias, which matters unusually here because the system prioritises signals and therefore determines which exposures surface and which do not. The company describes itself as an AI lab and publishes a Research page, which was not opened in this pass and is named here rather than credited, since a page title is not evidence of its contents. The agreement moves in the opposite direction, recording at section 8.3 that the company has no control over the output of the AI features. Searched the home page and the terms of use on 4 September 2026.

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.

A generic security clause covers the product without addressing what happens to customer content after processing. Section 7.3 commits to reasonable industry-standard technical and organisational security measures during the subscription term, with no control named: no encryption standard, no access model, no logging, no testing cadence. There is no security page anywhere in the site navigation or footer, which is itself unusual at this customer tier. Retention is not addressed at all and section 10.4 provides only that after termination the company may delete Customer Data, which is permissive and sits against the perpetual licence granted at section 5.1. Sub-processors are acknowledged to exist at section 8.1 and none is named. No incident or breach notification practice was located. The privacy policy is incorporated by reference at section 7.1 and was not opened in this pass; section 7.1 is headed Privacy Policy and DPA but no data processing agreement is linked or published.

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.

A real published position, short of the full picture. Section 14.1 gives a company indemnity, narrow but genuine, covering third-party claims that the grant of the right to use the service infringes a valid United States patent, with five named exclusions and remedies at section 6.4, and the agreement states expressly that these are the sole and entire liability and exclusive remedy for intellectual property infringement. Section 13 caps aggregate liability at fees paid in the preceding twelve months, dropping to 50 US dollars for trial and pre-released services under section 2.3 and for the drug and device intelligence trial. What is absent is any warranty a buyer could invoke: section 12.1 disclaims accuracy, completeness, availability and security outright, section 8.3 states the company shall not be liable for the performance of the AI features and does not guarantee the accuracy or reliability of any output, and the trial terms disclaim liability for undetected or mischaracterised signals, risks or cases. No insurance position was located.

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.

No integration into practice systems was located. Section 8.4 of the agreement refers to integration with certain third-party applications as set forth in the company's documentation or within the service, and then spends the rest of the clause disclaiming responsibility for them, without naming a single one. No case management system, document management system, intake platform or CRM appears anywhere on the surfaces read, and there is no integrations page or developer documentation in the navigation or footer. This matters for the buyer segment named first on the site, since a plaintiff firm receiving case signals needs them to land in an intake or matter system. The Platform and Law Firms pages were not opened in this pass and are named here so the limit is visible. Checked 4 September 2026.

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.

Nothing is published about where the software runs or where customer content sits. No region, country, hosting provider or data centre is named on any surface, and no tenancy model is described: nothing states whether the platform is single or multi-tenant or how one customer's data is separated from another's. The agreement says only, at section 8.1, that the service is hosted and made available by certain sub-processors, none of them named. No deployment option, private tier or residency commitment of any kind is offered. The one geographic signal is indirect, in the export control clause at section 15.1 referring to Israeli, United States or foreign export controls. Searched the home page, the terms of use, the site navigation and the footer on 4 September 2026.

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.

No independent security attestation was located. No SOC 2, ISO 27001, HIPAA or any other framework is claimed anywhere, no auditor or certifying body is named, no report is offered at any access tier, and no penetration testing is mentioned. There is no security page and no trust centre: the site navigation carries Law Firms, Insurance, Compliance, Research and a Company menu, and the footer carries only a cookie policy, a privacy policy and the terms of use. The agreement offers section 7.3 alone, promising reasonable industry-standard measures without naming a standard. There are no unsupported badges on display either, which is why this sits at the floor on absence rather than on overclaiming. Checked home page, terms of use, site navigation and footer on 4 September 2026.

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.

Partial disclosure, with the providers named and change notification expressly refused. Section 8.3 identifies the AI service providers as third parties including Anthropic Claude and OpenAI, both rendered with spelling errors in the published text, and makes clear that the AI features are mandatory and integrated by default. That is more than most records in this corpus publish, and it appears in the agreement rather than in marketing. The same clause then removes the second limb: the company reserves the right to use different AI service providers at any time without providing prior notice, and a customer who objects to a current or future provider has no remedy but to stop using the service. The word including leaves the list non-exhaustive. No specific model or version is named, no location is given for inference, and no subprocessor list is published, with section 8.1 acknowledging sub-processors without identifying any.

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 in the navigation or the footer, no tier names, no feature split, no indication of whether charging is per user, per matter, per signal or per case, and no free trial or self-serve route. Every call to action resolves to requesting a demo, talking to the team or booking time. What the agreement reveals is the shape of the arrangement rather than its price: fees are set in an Order Form, stated in US dollars, payable annually in advance, non-cancelable and non-refundable, with automatic renewal unless cancelled 30 days before expiry and overage fees for excessive use. A banner stating that 78 per cent of production is already reserved suggests a capacity-constrained commercial model, which makes the absence of any published unit more significant rather than less. No VendorPricing row is written.

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.

Segment coverage is described with substance across three distinct buyers and the boundaries are left open. Each segment carries its own page and its own stated job: law firms identifying emerging litigation earlier and building case portfolios, insurers detecting exposure and pricing risk, and corporate compliance teams surfacing exposure across their external digital footprint. That is an unusually clear statement of who the product is for. Practice breadth is quantified as more than 50 legal exposure categories covered, and a Domain Coverage page exists in the footer, which was not opened in this pass. The named customer set corroborates the primary segment, being overwhelmingly plaintiff and class action firms. What is missing is the edge: no individual practice area is named on the surfaces read, no firm size is addressed, no jurisdiction of coverage is stated, and nothing says which exposure types or industries fall outside the system.