S

Syllo

The models drive the capabilities the vendor leads with, on a platform whose conventional layer would still function without them, which is the B band. Agentic Issue Tagging coordinates multiple language models in distinct roles to run first and second level review with unlimited issue codes, and the open prompting engine answers questions across the case record and drafts summaries, deposition digests and chronologies.

The vendor describes itself as a Litigation AI platform and says it was founded to build the data and workflow infrastructure into which language models are woven. Beneath that sits a full eDiscovery module that stands on its own: native and load file processing with deduplication, threading and family grouping, Boolean and metadata search, multi-panel review with bulk redaction, Bates stamping, load file generation and privilege log automation, plus case management covering chronologies, exhibits, binders and transcript designations.

Strip the models out and a litigation support team still has a review and case management platform, which is why this is B rather than A. The Litigation Agent, which would push further toward the A description, is in early access and not credited. Verified 18 September 2026.

Measured accuracy is published for document classification and grounding is documented for generated output, short of measurement of the generated output itself, which is the B band. The white paper of 21 March 2025 reports that across the last ten completed responsiveness reviews in live litigation the lowest estimated recall was 93.4 per cent, the average 97.8 and the median 99.4, with median estimated precision of 85.9 per cent and average 79.7, and it gives per-matter figures with dataset sizes and the firm's own precision and elusion testing as the method.

Every tagging determination carries an issue-specific rationale pinpointing the text that triggered it, and answers from the prompting engine come with hyperlinked citations to the case record. Three things hold it below A. The measurements cover responsiveness classification, not the accuracy of generated answers, summaries or digests, for which nothing is published. The failure modes the paper names are those of competing approaches, prompt overload, context windows and hallucination in linear review, rather than this system's own.

And nothing states what the prompting engine does when the record does not support an answer. The benchmark is eighteen months old at this date. Grounding is to the customer's own record rather than to primary law, so citation status checking does not arise. Verified 18 September 2026.

Real review surfaces and a stated validation route, short of a control structure with a threshold or a categorical constraint, which is the B band. The white paper describes the review system as coordinating multiple language models that decide how to conduct the review within guidelines set by users. The product page states that case teams validate the outcomes of the agentic issue coding to ensure defensibility for outgoing productions, every determination carries a written rationale pointing to the triggering text, tag conflict detection is built into quality control, and the white paper's case studies show the validation in practice as precision testing on tagged sets and elusion testing on the null set.

Customer and Case Administrator roles in the terms control who can act on a case. What the A band asks for is not published: no threshold at which the system stops or escalates, no mode or tier carrying a stated constraint on what its output may be used for, and nothing on what happens after a coding error reaches a production. The home page headline claims the platform can autonomously manage any matter end to end; that describes the Litigation Agent, which is in early access, and no oversight structure is published for it. Verified 18 September 2026.

Named firms, a date, figures for what changed and a method a reader can assess, which is the A band. The white paper of 21 March 2025 lists practitioner contributors from seven firms, Ballard Spahr, Mayer Brown, Nixon Peabody, Outten & Golden, Pillsbury Winthrop Shaw Pittman, Quinn Emanuel and Royer Cooper Cohen Braunfeld, and reports matter-level results measured by the firms' own teams. Ballard Spahr's trial team ran precision and elusion testing on a review of more than 100,000 documents against more than 25 issue codes and found estimated precision of 95.56 per cent and recall of 99.4 per cent, after about three hours of set-up.

Royer Cooper Cohen Braunfeld ran a head-to-head on a random sample of just under 16,000 documents from more than 150,000: the managed review team using continuous active learning reached estimated recall below 67 per cent at over two dollars a document, and the system reached recall of 93.44 per cent and precision of 69.81 per cent. Outten & Golden reports precision of 84.09 per cent and estimated recall of 100 per cent on 12,543 documents with 28 tags, and Quinn Emanuel a zero-shot review of more than 40,000 documents at 98.69 per cent recall and 92.83 per cent precision.

The limits are stated here so the grade is read correctly: the paper is vendor published, the matters are not named, practitioner contributors wrote in their individual capacities, and the paper's own footnotes disclose that one technical contributor is a research advisor and one practitioner a product advisor to the vendor. Verified 18 September 2026.

Substantive written commitments, including express treatment of privilege and work product, defeated on training, which is the B band. The Legal Disclosure and Subpoena Policy, incorporated by the terms at section 9, states that much of what users store is confidential and privileged under the attorney-client privilege and the work product doctrine, argues that the customer rather than the vendor is the proper target of legal process, and commits to written notice so a customer can seek a protective order.

The terms add that Content is encrypted in transit and at rest, that the vendor acquires no rights beyond those needed to offer the service, that Content is not shared with third parties except service providers, legal process and security monitoring, that vendor personnel may not view Content without permission or a support, security or legal reason under internal controls (sections 13.1 and 16.6.2), and that Content may not be used outside provision of the service (16.6.4).

Segregation is documented at case level through Case and Customer Administrators, and the product page states matter-level isolation is standard. What fails the A band is training: section 8.4.1 licenses Content for providing or improving the service, and no commitment against model training is published anywhere. The position of any third-party model provider is not stated. The customer Licensing Agreement, which prevails over the terms, is not published. Verified 18 September 2026.

The terms place professional responsibility on the user and widen the audience, while the marketing describes the product in terms of legal analysis and strategy, which is the C band. Section 5.4 of the terms states that the service is not intended for use solely by attorneys, judges, paralegals or others in the legal profession and may be used by anyone holding a licensing agreement, and that the vendor has no responsibility to determine or monitor whether a user is engaged in the unlicensed practice of law.

Section 5.8 makes each user responsible for ensuring they are authorised to practise where they use the service for the practice of law, and section 15 disclaims any warranty that content provided through the service is accurate. Against that, the product is marketed for work product generation, legal analysis and case strategy, and the home page headline speaks of autonomously managing a matter. No statement says the product does not give legal advice, nothing addresses a supervising lawyer's competence or supervision duties, and no jurisdictional limit is stated beyond a US-only account rule that the terms tie to privacy and data protection law rather than to practice. Verified 18 September 2026.

An evaluation practice is published without any governance framework, owner or pre-release testing regime around it, which places this nearest the C band. What exists is real: the white paper describes how outputs are validated in live matters, through precision testing of tagged sets and elusion testing of the null set, reports aggregated results across ten reviews, and names a Carnegie Mellon Language Technologies Institute researcher as a technical contributor.

The work with Syllo page says the AI systems are designed for rigorous human review and feedback. The terms separate Beta Previews from the audited service, stating early access features are not subject to the same security measures and auditing. None of that is governance in the band's sense. No responsible AI policy or principles page was located, no individual or function is named as accountable for model behaviour, nothing describes what is tested before a model or feature is released, and uneven output across matter types or parties is not addressed.

The band's C description speaks of principles without a mechanism, and this record is closer to the reverse, a measurement practice without stated principles; C is the nearest fit. Verified 18 September 2026.

Substantive published controls across most limbs, short of incident practice and a processor list covering the AI, which is the B band. The security section of the Work with Syllo page states encryption in transit and at rest in line with NIST SP 800-57, automatic backups, regionalised storage, continuous vulnerability scanning and penetration testing, and multi-factor authentication. The terms name where Content is stored, Amazon Web Services virtual servers and Amazon S3 controlled by the vendor and ElasticSearch for indexing (section 13.3), restrict personnel access to Content (13.1), require single-person credentials and user notice of a compromise within one business day (2.1), reserve deletion of a cancelled user's personal knowledge bases within 90 days with residue in encrypted backups (2.3.2), and aim for 30 days' notice before a feature is discontinued so users can export (16.3).

Three limbs fail. No commitment to notify customers of a security incident was located. No retention period is stated for prompts, outputs or case content during or after a subscription, and section 2.1.8 places export on the user with no obligation on the vendor after termination. No list names who processes content for the AI features. The product Privacy Policy and Data Security Policy, which the terms place at app.syllohq.com/policies, return an empty page to automated retrieval and could not be read on this date; they are the documents most likely to move this row. Verified 18 September 2026.

The published terms address liability only through a disclaimer, a nominal cap and an indemnity running to the vendor, which is the C band. Section 15 provides the service on an as is, as available basis, disclaims any warranty that content provided through it is accurate, reliable or correct, excludes all categories of damages, caps the vendor's total liability to an end user at one hundred dollars, and excludes liability for breaches at third-party providers.

The indemnity runs the other way: each user indemnifies and defends the vendor for their use, their content and their breaches. Nothing addresses AI output specifically, no vendor indemnity is offered, and no insurance position is published. The terms bind end users and state that the customer's Licensing Agreement prevails where the two conflict, so the terms a firm actually negotiates may differ, but that agreement is not published and a buyer cannot read it before a sales process. Verified 18 September 2026.

The product stands alone by design, which is the D band read literally. The vendor argues on its own platform page that API integration and acquisition have both failed to solve fragmented litigation tooling, and positions a single environment holding the whole record as the answer. What it supports is interchange rather than integration: native file and load file ingestion with deduplication, threading and family grouping, load file generation and Bates stamping at production, and import of standard court-reporting formats in text, audio and video.

Those are the formats data arrives and leaves in, and they mean moving a matter onto the platform is routine, but no connection to any external system is named. No document management system, no Microsoft Word or Outlook add-in, no review platform, matter management, e-billing or court filing integration was located on the surfaces read, and no API or developer documentation is published. The grade records a stated design choice, not a gap the vendor has left unexplained. Verified 18 September 2026.

Residency is offered and the hosting provider is named, with the deployment options themselves left unspecified, which is the B band. The security section states that multiple secure deployment options are available, that data storage is regionalised based on customer needs, and that for firm-level engagements the vendor will work with a firm's security team to customise deployment to its requirements. The terms name the infrastructure: Content is stored on Amazon Web Services virtual servers and Amazon S3 storage controlled by the vendor, with ElasticSearch for indexing (section 13.3), and user accounts may not be used outside the United States, a restriction the terms tie to privacy and data protection law (sections 2.1 and 5.6).

What is missing for A is specificity. The deployment options are not named, so whether single-tenant, private cloud or on-premises arrangements exist cannot be established, no regions are listed, and nothing distinguishes where model inference runs from where data is stored. Verified 18 September 2026.

A named attestation with stated scope and a named auditor, short of any route to the report, which is the B band. The compliance section states SOC 2 Type II certification in Security, Confidentiality, Availability and Privacy, compliance with the AICPA SOC standards for service organisations, and third-party validation by Prescient Assurance. Naming four of the five trust services criteria and the attesting firm is more than most records at this band carry.

What keeps it off A is access and currency: no trust centre was located on the site or by search on this date, no report period or issue date is given, and no request route, self-serve or otherwise, is described. Continuous vulnerability scanning and penetration testing are stated on the same page, without summaries. Verified 18 September 2026.

The architecture is described in unusual detail and no provider or model is named, which is the B band. Product pages state the platform is powered by a combination of proprietary Syllo models and foundational language models. The white paper goes further, describing an ensemble of varying-sized language models performing distinct roles, strategising, determining next steps, quality control, extracting learnings from documents, synthesising across documents, summarising and resolving disagreement between models, with telemetry routing harder material to higher-end models and simpler material to cheaper ones.

A buyer learns how the system is built. The buyer does not learn whose models these are: no provider, model family or version is named on any surface read, nothing states where inference runs as distinct from the AWS storage named in the terms, and nothing addresses what a model provider may retain. On change, the terms run the other way, reserving the right to modify any feature at any time with or without notice (section 13.7). Verified 18 September 2026.

No price, unit of charge, tier or inclusion list is published, which is the D band. Searched the home page, the Work with Syllo page including its pricing section, the Unified Platform and Agentic Issue Tagging pages, the Terms of Service and the white paper on 18 September 2026. The pricing section names two buyer segments and routes both to a sales conversation: in-house teams are offered the platform on an enterprise basis, and law firms on a firm-wide enterprise partnership.

Firm-wide describes the scope of a licence rather than a unit a buyer could price. The only figures on the estate are competitors' costs, the white paper reporting a managed review team at more than two dollars per document, alongside claims of review at a fraction of the time and cost. No pricing row is written. Verified 18 September 2026.

Segments and matter types are described with evidence behind them, short of the boundaries, which is the B band. The vendor names AmLaw 100 firms and boutique litigation firms, and in-house teams at corporations managing complex litigation portfolios, with a dedicated in-house page. The white paper evidences the breadth with named firms from AmLaw defence practices to a plaintiffs' employment firm, and lists matter types the system has been used on: antitrust, environmental, contract, employment, patent, bankruptcy, mass tort, construction, investment and shareholder disputes, M&A, automotive, real estate and insurance coverage litigation, across datasets from thousands to more than two million documents.

The scope is litigation throughout. What is left open is where it stops: nothing addresses government or court use, nothing names matter types or data types it handles poorly, and the one stated boundary, that accounts may not be used outside the United States, is framed as a data protection rule rather than a coverage limit. Verified 18 September 2026.