Theo Ai vs Turbo Law: how they compare in 2026
Theo Ai and Turbo Law are not usually weighed against each other: Theo Ai gives a corporate legal department a portfolio view of its litigation and recommends which matters to settle and which to fight, while Turbo Law is the tool defense counsel use inside one matter file to draft motions and evaluate the case. On one grid, both produce the same consequential number, an estimate of what a case is worth, and both leave the same things unpublished about it. Neither publishes an accuracy measure for its valuations, or the customer agreement that would say who bears the loss if one is wrong. Otherwise they sit far apart: Turbo Law sits in the top two bands on ten of fifteen axes and Theo Ai on two of fifteen. Turbo Law states that customer content does not train foundation models, that high stakes outputs go to attorney review, and that it holds SOC 2 Type II. Theo Ai's privacy policy excludes the enterprise data it processes, and no security attestation, oversight structure or training position was located for a product that reads a company's email, invoices and filings.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
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 artificial intelligence is the product, and there is nothing underneath it to sell, which is the A band. Every function the vendor publishes is model output: synthesising a caseload from documents collected automatically, extracting structure from emails and invoices, recognising patterns across a portfolio, projecting cost trajectories, flagging matters as under-reserved, summarising each case with attribution to source material, and recommending which claims to settle and which to fight. There is no repository, workflow engine, billing system or document store being sold underneath that a customer would still license if the models were removed; the vendor's own framing is a prediction engine, and its earliest published description was of a proprietary data model and prediction engine rather than of software with intelligence added. The 2024 to 2025 announcements describe the roadmap in the same terms, the funding being applied to expand proprietary data pipelines, enhance the legal corpus and reinforce supervised learning with legal experts. This grade says nothing about how well the models work or how much is disclosed about them; on that, see the Model Supply Chain and Citation Accuracy rows, both of which sit far lower. Verified 12 September 2026.
The models are the product, which is the A band. Turbo Law reads the whole matter file with AI to build its Fact Graph, and every module (review, drafting, research, strategy and settlement, and the assistant) generates work product from it: chronologies, contradiction detection, privilege flags, motions, discovery responses, case evaluations and settlement-value models. There is no underlying case management system that works without the models. Verified 18 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 claimed and nothing is measured on any surface the vendor publishes, which is C. The grounding claim is a single sentence and it is the only one located: every case is summarised with clear attribution to source material. That is a real architectural claim in the product's own idiom, and R40 keeps the D limb from firing because the accuracy content is not a bare assertion that the system does not hallucinate. Nothing supports it further. No accuracy figure, test set, error rate, evaluation or benchmark appears on any first-party surface, and no failure mode is named. Recorded and deliberately not credited: the chief executive is quoted in a December 2025 interview with a technology publication claiming roughly 85 per cent predictive accuracy against 60 to 65 per cent for human reviewers. That is a third-party publication rather than a vendor disclosure, it names no test set and no method, and it is not what R15 credited on Pre/Dicta, which published its figure, described a test set of more than 50,000 motions across 94 federal district courts and named its own limits. The contrast is the point: this corpus already holds a litigation prediction product that published its measurement, so the absence here is a choice rather than a feature of the product class. R15 also applies to the citator and authority limbs, which do not bite on a product that predicts outcomes rather than citing law. Verified 12 September 2026.
Grounding is real and described, and accuracy testing is claimed without figures, which is the B band. The security page says fact-based outputs are line-cited to a source document or the public record and that the platform will not state what it cannot cite, and that every drafting and analysis workflow is measured against vertical-specific accuracy standards before it ships; the product pages say every output is line-cited to the file and case-law search cites the relevant holdings. No accuracy standard, test set or result is published. Verified 18 September 2026.
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.
No oversight structure is published for a system that acts on live client matters, which is the D band. What the product does is published plainly: it recommends which claims to settle versus fight, automatically and in real time, flags matters as under-reserved, and in the vendor's words manages the minutiae. What happens around that decision is published nowhere. No mode structure, no confidence threshold, no escalation path, no review surface, no statement that a lawyer approves an output before it is relied on, and no account of what the system does when the evidence is thin. The two documents the vendor does publish do not fill the gap: the Terms of Service govern use of the website and never mention output, and the Privacy Policy expressly excludes information processed on behalf of enterprise customers. The nearest thing to a human-in-the-loop statement anywhere on the estate is incidental marketing copy, that the platform lets a team review the portfolio and make decisions with confidence and lets attorneys and paralegals focus on higher-level work. That positions a human as the reader of an output; it is not an oversight mechanism and it is not offered as one. This grade is a statement about what is published on the date shown, not a finding that no oversight exists in the product. Verified 12 September 2026.
A categorical review constraint on named high-stakes outputs, which meets the A band under R124(2). The security page says case evaluations, settlement recommendations and motion drafts surface for attorney review before anything goes external, and states the operating model as the platform proposes, the team verifies and partners decide; key conclusions are hyperlinked to source material for that verification. The assistant runs multi-step and scheduled tasks, but its output returns to the team as cited work product. No confidence threshold is published. Flagged for sampling. Verified 18 September 2026.
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.
What stands in for deployment evidence is an advisory board and an investor list, and the outcome figures published carry no basis, which is C. The site's principal social proof is a General Counsel Board of sixteen named senior legal figures, one identified on the page as the General Counsel of SmartBiz Bank, alongside a strip of investor logos. An advisory board is not a customer and is recorded here rather than credited; nor are investors. No customer is named anywhere on the current estate, no case study is published, and no logo wall of users appears. One named deployment does exist and is first-party: the vendor announced a partnership with Mustang Litigation Funding in March 2025, with the counterparty's chief technology officer quoted, but the account is forward-looking rather than measured, projecting that a one per cent annual improvement in case selection would raise returns. The figures that circulate are all unattributed or third-party: an early claim that review time falls from between seven and thirty days to seconds, an assertion that early users found the platform verified their underwriting teams' conclusions, and the 85 per cent accuracy claim made in an interview. None states a basis. The record would move on a named customer with a figure and a method, which is the ordinary route out of this band. Verified 12 September 2026.
Outcome claims without named customers or method, which is the C band. Turbo Law reports more than 1,800 active matters on the platform and says firms using it report a 60 per cent reduction in non-billable hours per matter, 10 per cent fewer write-offs and profitability on alternative fee arrangements; no customer is named on the surfaces read and no method is given for the figures. Verified 18 September 2026.
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 is published on how client confidences are handled by a product built to ingest them, which is the D band, and the shape of the gap is worth stating precisely because it is not an unread document. Two instruments are published and both were read in full. The Terms of Service, last updated August 2025, govern use of the website at theoai.ai; they contain no confidentiality obligation running to a customer, no security commitment, no data handling terms and no reference to case material. The Privacy Policy, last updated 6 April 2025, opens by excluding the very thing this axis asks about: it states that it does not apply to information processed on behalf of enterprise customers as a processor or service provider, and that such use is governed by agreements with those customers. Those agreements are not published. So a buyer evaluating this product before signing can read nothing about training on client data, segregation between customers or matters, retention or deletion of case files, or what any model provider may retain. That matters more here than on most records, because the published product ingests litigation files automatically from a customer's own Microsoft, Google, Workday and Salesforce systems. Recorded and not credited: the application's sign-in page offers single sign-on, which is an access method rather than a confidentiality commitment. Verified 12 September 2026.
A substantive public position that addresses privilege, with the customer agreement unpublished, which is the B band. The privacy policy says Turbo Law does not use customer content to train or improve foundation models, that any fine-tuning happens only at the customer's direction, that third-party AI providers are contractually barred from training on customer data, that Turbo Law does not assert or waive privilege or work product on the customer's behalf, and that staff access is limited to authorised purposes; the security page adds tenant isolation that its own team cannot cross without a support ticket and client approval. The security page says the no-training commitment is enforced in contracts, but the customer agreement and DPA are not published. Verified 18 September 2026.
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. 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 is published on the advice line for a product that produces legal work, which is the D band. The product tells a legal team what a case is likely to be worth and recommends whether to settle it or fight it. No statement was located anywhere on the estate that the vendor is not a law firm, that its output is not legal advice, that a lawyer must exercise independent professional judgement before relying on it, or that the recommendation is a suggestion rather than a conclusion. The two published documents do not supply it: the Terms of Service disclaim warranties in respect of the website and its content and say nothing about advice, professional responsibility or the practice of law, and the Privacy Policy is a data document that expressly does not reach the product. Nothing addresses who may use the platform, whether a supervising lawyer's competence duties are engaged when a settlement recommendation is acted on, or any jurisdictional limit, and the site is marketed to litigators, paralegals and legal operations staff alike without distinguishing between them. This grade describes the published record on the date shown; the absence is unusually complete for a product in this category, and it is the row most likely to move if a customer agreement is ever published. Verified 12 September 2026.
A clear statement that the product is software and not legal services, with the attorney's role stated, which is the B band. The Terms of Use say Turbo Law provides software, not legal services, that nothing creates an attorney-client relationship, and that legal decisions should be made by licensed attorneys exercising independent professional judgement; the security page says the platform is built to make litigators faster, not to practise law. The buyers are law firms and insurance carriers. No jurisdictional limits are stated. Verified 18 September 2026.
AI Governance and Bias Disclosure
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.
No governance position is published for a system whose output affects legal outcomes, which is the D band. There is no responsible AI page, no set of published principles, no framework such as the NIST AI Risk Management Framework, no certification, no accountable person or function named, no description of what is evaluated before a model change ships, and no statement of any kind about how the system is tested. Neither published document addresses AI: the Terms of Service govern website use and the Privacy Policy excludes enterprise customer processing. The bias question is live on this product in a way it is not on most of the corpus and its absence is therefore worth naming rather than passing over. A model that predicts settlement values and case outcomes from historical litigation data is learning from a record that reflects who was represented by whom, in which venue, against which counterparty, and what those parties could afford to accept; nothing published addresses whether output varies by claimant characteristics, venue, counsel or claim type, or whether that has been examined at all. The vendor's own published account of its method, that it reinforces supervised learning with legal experts, is a training-process description rather than a governance disclosure and is recorded on the Model Supply Chain row. Verified 12 September 2026.
A stated release gate with substance, short of testing results, which is the B band. The security page says every drafting and analysis workflow is measured against vertical-specific accuracy standards before it ships and does not ship if it fails; the Trust Center lists board oversight briefings, annual risk assessments and quarterly critical system reviews. The standards, results and any assessment of uneven performance are not published. Verified 18 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.
Nothing is published on retention, deletion or access for a system that holds client documents, which is the D band, and the absence is established rather than untested. The Privacy Policy carries a retention section, and it is expressly scoped away from the product: the policy states at the outset that it does not apply to information processed on behalf of enterprise customers, so its retention language, which in any event says only that personal information is kept as long as needed or permitted, reaches website visitors rather than case files. Its security section is a single sentence about reasonable organisational, technical and administrative measures, with no encryption commitment, no access control description and no incident practice. There is no subprocessor list, no security page, no trust centre, no data processing addendum and no statement of where anything is hosted. Deletion is addressed nowhere for customer material. Two observations are recorded rather than credited, so a reader knows what was found: the application sign-in page offers single sign-on alongside password authentication, which is an authentication method and not a stewardship disclosure; and the Terms of Service grant a perpetual, irrevocable, sublicensable licence over user Submissions, which on its own words covers material posted through the website rather than case documents uploaded to the application. Verified 12 September 2026.
Real published practices with the contractual detail withheld, which is the B band. The security page describes tenant isolation, staff access only with a support ticket and client approval, encryption at rest and in transit, enforced two-factor authentication, exportable logs of every access, query and export with one-year retention by default, and HIPAA business associate agreements; the privacy policy says the subprocessor list is available from the account team. Deletion and retention of customer content are left to the unpublished customer agreement, and the Trust Center's answer on incident notification is collapsed and could not be read. The privacy page carries two different policy texts, one covering the platform and one covering the website only. Verified 18 September 2026.
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.
Liability is addressed only through a standard limitation clause that disclaims the exposure the product creates, which is C, and here the clause reaches the product by an indirect route worth setting out. The application's sign-in page states that by signing in the user agrees to the Terms of Service, and those Terms are a website terms of use. Their liability provisions are nonetheless what a user is bound by. Section 14 supplies the site and its contents as is, where is and where available, without warranties of any kind, and disclaims merchantability, fitness for purpose, non-infringement and title. Section 16 excludes indirect, incidental, consequential, special, exemplary and punitive damages including loss of profits, use and data, makes ceasing to use the site the user's sole and exclusive remedy for dissatisfaction, and caps aggregate liability at the total amount, if any, paid to the company to use the site. Section 17 runs an indemnity the other way, the user defending and indemnifying the company for claims arising out of their use of the site. Nothing anywhere allocates the loss when a settlement recommendation or an exposure figure is wrong, no indemnity covers output, and no insurance or warranty position is published. Verified 12 September 2026.
No published position on what the vendor stands behind, which is the D band. The only published terms govern the website, cap liability for the site at $100, and say the platform is governed by separate customer agreements, which are not published. Nothing located addresses warranties, indemnities, caps or remedies for the platform or its AI output. Verified 18 September 2026.
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.
Integrations are named in a sentence with no documentation an implementer could use, which is C. Four systems are named and they are the right ones for the workflow: the vendor states that the platform gathers data points automatically from the tools a team uses every day, from Microsoft, Google, Workday and Salesforce, and adds that the emails and invoices held in them become a source of intelligence. That is more than a logo wall and it does describe, at the highest level, what moves. What is absent is everything below that level. No integration page exists, no connector list, no developer documentation, no API, no authentication or configuration requirement, no statement of direction of flow, no description of what is read as against written back, and nothing about what a customer must do to connect a system. The one technical fact established directly is on the application sign-in page, which supports single sign-on alongside password authentication; the identity provider is not named. Recorded and not credited because it is third-party: technology press reports that the product is designed to work inside email rather than through a dashboard, which if published first-party would be a material integration fact. Verified 12 September 2026.
Integration is claimed without named systems, which is the C band. Press material says the platform plugs into existing case files, and a forward-deployed team stands it up inside each firm; no document management, case management, e-billing or carrier system is named on the surfaces read, and PACER and docket monitoring are listed as coming soon. Verified 18 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 on where the software runs or where client data sits, which is the D band. No hosting arrangement, cloud provider, region, data centre, tenancy model or residency option is stated on any surface located. The Terms of Service address jurisdiction only in the sense of governing law and market availability, stating that the site is controlled or operated from the United States, that it is not intended to subject the company to any non-United States jurisdiction, that it may not be appropriate or available in some non-United States jurisdictions, and that the company may limit availability to any person, geographic area or jurisdiction it chooses. That is a statement about who may use the website, not about where customer data is processed or stored, and it is recorded here so a reader can see that the nearest published geography is not residency. The Privacy Policy would be the ordinary home for a processing-location statement and it expressly excludes enterprise customer information from its scope. The product is delivered as a hosted web application at an app subdomain, which establishes that it is cloud-delivered and nothing more. Verified 12 September 2026.
The host is named without regions or options, which is the C band. The security page says the application and data run in dedicated Microsoft Azure environments with tenant isolation; the Trust Center's answer on where data is stored is collapsed and could not be read, and no region, residency option or processing location for AI calls is published. Verified 18 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, which is the D band. There is no trust centre, no security page, no compliance page, no certification badge of any kind on any surface read, and no reference anywhere to SOC 2, ISO 27001 or any equivalent standard, whether as held, in progress or available on request. No penetration testing, vulnerability disclosure programme or audit regime is described. There is accordingly no access flow to grade under R5, because there is nothing to request: the estate contains no sales-gated security artifact, not even a contact route for a security review. The only security statement located anywhere is one sentence in the Privacy Policy, that the company seeks to use reasonable organisational, technical and administrative measures to protect personal information within its organisation, followed by the standard acknowledgement that no system can be guaranteed secure; and that policy states at its outset that it does not apply to information processed on behalf of enterprise customers. The ordinary reading of this record is early-stage rather than evasive, and the note should be read that way, but the axis grades what a buyer can verify before signing and the answer is nothing. Verified 12 September 2026.
A named standard with a readable trust centre, short of dated reports, which is the B band. The Trust Center at trust.turbolaw.ai lists SOC 2 Type II and HIPAA as compliant, names Prescient Security, sets out control families and more than 150 documents available on request, and the security page says SOC 2 documentation is available under NDA. The security page describes controls as aligned to the Trust Services Criteria and continuously monitored; no audit period, report date or auditor opinion is published. Verified 18 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.
The vendor refers to proprietary models without identifying what sits underneath, which is C in its plainest form. The first-party language is consistent across the estate and the announcements: a proprietary data model and prediction engine, proprietary data pipelines, an enhanced legal corpus, supervised learning reinforced with legal experts, and firm-specific prediction engines built on a customer's own case history. Every one of those describes an approach rather than a component. No model is named, no version is given, no third-party provider is identified, no hosting arrangement for inference is stated, and no commitment exists to notify a customer if any of it changes. There is no subprocessor list to fall back on. Recorded and expressly not credited, because it is third-party rather than vendor disclosure: technology press reporting in late 2025 states that the platform draws on multiple large language models from OpenAI, Google and Anthropic and is built primarily on Google's Gemini, fine-tuned for vertical legal tasks. If the vendor published that itself it would move this row materially, which is exactly the point of the axis. What a customer inherits today is undisclosed, on a product whose whole function is to send case files, and by the vendor's own account medical and insurance records, to a model. Verified 12 September 2026.
The model architecture is described without naming providers, which is the B band. The security page says Turbo Law routes each workflow to the model best suited to it and changes models as the state of the art changes, with no single-vendor dependency, on dedicated Azure infrastructure; the privacy policy says third-party LLM providers are contractually barred from training on customer data. No model or provider is named, and no notice of model changes is committed. Verified 18 September 2026.
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, which is the D band. There is no pricing page, no plan or tier name, no band, no minimum, no term, no statement of whether the product is licensed per user, per matter, per portfolio or per prediction, and nothing about what implementation or data onboarding adds. The site's only commercial surfaces are a waitlist form and a calendar link to book a demonstration. The Terms of Service contain generic transactional boilerplate about products, orders, shipping and credit cards that appears to be unadapted template language rather than a description of how this product is sold, and it is recorded as such rather than read as pricing structure. Under R10's closing discipline a page that only invites a conversation is an absence and belongs in this note alone, so no VendorPricing row is written for this record. One adjacent fact is recorded because it bears on what a buyer can obtain rather than on what it costs: the site states that onboarding is currently by invitation only, so the commercial route is not merely unpriced but gated. Verified 12 September 2026.
Nothing about pricing is published, which is the D band. The site has no pricing page and routes every buyer to a demo request; neither the unit, the structure nor any figure is stated. Verified 18 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.
Coverage is described with real substance and the boundaries are left open, which is B, and this is the strongest of the disclosure rows on the record. The buyer is identified precisely rather than broadly: litigation defence teams inside corporate legal departments, with two published solution surfaces, one for the general counsel and one for product liability, and named user roles of attorney, paralegal and legal operations. Sector coverage is enumerated rather than claimed at large, naming retail, insurance, chemicals, hospitality and gaming including hotels, resorts, casinos and cruise ships, and pharmaceuticals and medical devices, with the explanation that these are the sectors where litigation volume never slows. Matter coverage is the most specific element and is unusual for a vendor this size: a published table sets out six categories the product is built for, general liability claims, workers' compensation and employee claims, employment and human resources disputes, regulatory and compliance matters, customer and vendor disputes, and internal investigations, each with indicative annual matter counts and page volumes. What is left open holds it off A. No jurisdiction is named anywhere, nothing states which matter types or claim values fall outside the model's competence, and the historical shift in audience, from litigation funders to plaintiff-side selection to defence-side portfolio management, is not reconciled on the current estate. Verified 12 September 2026.
Segments and matter types are set out with substance, short of stated limits on the AI, which is the B band. Turbo Law sells to defence law firms and insurance carriers and names medical malpractice, mass tort, toxic tort, transportation, mergers and acquisitions, business litigation, employment, construction and real estate, and says it builds one litigation vertical at a time; drafting follows the jurisdiction's local rules and case-law search covers federal and state courts. Where the models are weaker is not stated. Verified 18 September 2026.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
No customer agreement is published and no policy states a position on training, so a client has no term to hold this vendor to. What is published, and both documents were read in full, is a website Terms of Service last updated August 2025 and a Privacy Policy last updated 6 April 2025, reachable as PDFs linked from the application sign-in page rather than from the marketing site, whose own privacy link is a dead anchor.
Neither reaches customer case material. The Terms govern use of the website at theoai.ai and never mention training, models or output. The Privacy Policy states in its second paragraph that it does not apply to information processed on behalf of enterprise customers as a processor or service provider, and that such use is governed by agreements with those customers; those agreements are not published anywhere located.
Two provisions are recorded so a reader can see what was considered and why neither carries the row. The Privacy Policy lists improving and developing products and services among its purposes, including data analysis and developing new services, but only for the website information within its scope. And Terms section 9 grants a perpetual, irrevocable, sublicensable license to use, analyze and exploit user Submissions for any purpose, which on its own words covers material posted through the website rather than case files uploaded to the application, though the application's sign-in flow binds users to those Terms.
The privacy policy and security page say customer content is never used to train shared, public or foundation models, by Turbo Law or any model vendor, and that any fine-tuning on customer content happens only at the customer's direction. The security page says this is enforced in contracts, but the customer agreement is not published, so no matching term was located.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
No located public material states how long case material, prompts or generated output are retained, and this is an established absence rather than an unfinished search. The Privacy Policy does carry a retention section, and it is scoped away from the product before it begins: the policy applies to information collected through the vendor's websites, social media pages and online communications, and expressly not to information processed on behalf of enterprise customers.
Within that scope it says only that personal information is retained for as long as needed or permitted in light of the purposes for which it was obtained, unless a longer period is required or permitted by law, which is a criterion rather than a period. The Terms of Service address retention nowhere. No data processing addendum, security page, trust center or subprocessor list exists on any surface located. Nothing states whether a case file uploaded for prediction is held after the prediction is produced, whether a customer can delete it, or what happens to anything at the end of a relationship.
The surfaces read on the date shown were the home page in full, the application sign-in page, and both published documents in full.
Retention is mentioned without a period for customer content or AI inputs and outputs. The one-year default sits under the audit-log heading on the security page, and the privacy policy says retention and deletion of customer content are governed by the unpublished customer agreement.
Ethical Walls and Matter Segregation
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. Nothing published describes separation between customers, between matters inside one customer, or between the users of a single deployment, and no permission model, role structure or access group is described anywhere. The one access-related fact established is on the application sign-in page, which supports single sign-on alongside password authentication; that is an authentication method rather than a segregation model and it is recorded rather than credited.
Two features of the published product make the silence worth naming. The vendor states that the platform collects data automatically from the systems a legal team already uses, naming Microsoft, Google, Workday and Salesforce, which means it reaches across a customer's own estate rather than sitting behind a single upload; and the announced roadmap is firm-specific prediction engines that leverage a customer's own case history, which raises directly the question of whether one customer's history can influence another's prediction, addressed nowhere.
The Privacy Policy that would ordinarily carry a processing-boundary statement expressly excludes enterprise customer information from its scope.
Tenant isolation between customers is claimed; walls within a customer are not yet available. The security page lists firm, practice-group and matter-level ethical walls under a coming-soon heading, and one version of the text says they are on the way. Roadmap items are not graded.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Disclosure to authorities is addressed and customer notice is absent, which is this value. The Privacy Policy's disclosure table lists authorities and others among the recipients and states that the vendor may disclose information to law enforcement, public, regulatory and government authorities, courts, or other entities obtaining valid subpoenas, where required by law or legal process or where the vendor deems disclosure appropriate for other legal, compliance or fraud-prevention purposes.
That second limb is broader than compulsion and is quoted so a reader sees it. Nothing anywhere commits the vendor to notify a customer of a demand, reserves discretion over whether to notify, promises to seek a protective order or to redirect the requester, or sets any timeframe. No transparency report exists. One scope limit belongs in the reading and does not change the value: the policy states that it does not apply to information processed on behalf of enterprise customers, so what a buyer is being told about here is website and business-contact information rather than the case files the product ingests, and for those the position is not addressed at all because no customer agreement is published.
The privacy policy says information may be disclosed to comply with law or protect rights, with no commitment to notify the customer first. The customer agreement, which may address this, is not published.
Primary Law Corpus Provenance
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, and on this record that is the sharpest of the disclosure gaps, because for a prediction engine the corpus is the product. The vendor refers to it repeatedly without ever identifying it: proprietary data pipelines, an enhanced legal corpus, a proprietary data model, supervised learning reinforced with legal experts, and firm-specific engines leveraging a customer's own case history.
Not one of those names a source, a jurisdiction, a court record set, a date range or a licensing basis, and no coverage statement of any kind appears. Two things a buyer would need to weigh are therefore unavailable. Whether the historical settlement and outcome data behind a prediction was lawfully obtained and on what terms, which matters because settlement figures are typically confidential; and whether a customer's own case history, once ingested to build a firm-specific engine, remains theirs, which is unaddressed because no customer agreement is published.
Recorded and not credited because it is third-party: technology press reporting in late 2025 describes the platform as built on private settlement records obtained through litigation funders. Verified against the home page, the application sign-in page and both published documents on the date shown.
Searched the product overview, security page, Trust Center, privacy policy and newsroom on 18 September 2026. Case-law search covers federal and state courts and settlement modeling uses jurisdictional outcomes, but the sources and licenses behind them are not named.
Good Law Verification
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, and on this product class the question does not arise in its usual form. The product cites no cases, statutes or regulations to a reader. It works on a customer's own matters and on historical outcome data, and what it emits is a prediction, an exposure figure, a reserve flag and a case summary, none of which is a proposition of law that a lawyer would shepardise.
The nearest adjacency is currency of a different kind, whether the historical data behind a prediction is recent enough for the prediction to hold as law and practice shift, and nothing published addresses that either: no update cadence, no data vintage and no statement of when a model was last retrained appears anywhere. Recorded so the row states the position rather than leaving a reader to infer it from a product category. The home page, the application sign-in page and both published documents were read on the date shown.
Searched the same surfaces on 18 September 2026. Case-law search returns citations to holdings, but no check of whether cited authority remains good law is described.
Refusal and Uncertainty Behavior
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. Nothing states that the system declines to predict, marks a prediction as unsupported, reports that a case falls outside the data it holds, or escalates rather than answering, and no evaluation demonstrating any such behavior is published. One distinction is worth drawing because it could otherwise be mistaken for a confidence signal.
This product's output is itself probabilistic: the vendor describes estimating the odds of success and a range of recovery, and a probability attached to an outcome is a statement about the case, not a statement about how much the model trusts its own reading of the file. Nothing published exposes the second thing. The gap matters on a product that recommends settling or fighting automatically and in real time, because the case where the system has too little to go on is exactly the case a reviewer needs flagged.
The surrounding documents do not fill it: the Terms of Service disclaim warranties as to the website's content, and the Privacy Policy does not reach the product.
The vendor states a behavior rule: fact-based outputs are line-cited to the file or the public record, and a claim that cannot be cited is not made. No confidence signal shown to the user is described.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
Searched on 12 September 2026, on both the product name and the company name, against published trackers of decisions on AI-generated fabricated citations including coverage of the Damien Charlotin AI Hallucination Cases database and two independent sanctions trackers, for any court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product. None located. This is a statement about the public record on that one subject as of the date shown, and under R119 this signal records fabricated citations and nothing else, so it is not a litigation history.
One note for a future reader: the vendor's name collides with several unrelated products and companies using the word Theo, so any future search needs the corporate qualifier to be meaningful. Note also that this product does not generate citations to authority, so the shape of exposure this signal tracks is not the shape this product presents.
Searched the AI Hallucination Cases database maintained by Damien Charlotin and trade press reporting on 18 September 2026 for court records addressing fabricated or hallucinated content in output from Turbo Law. None located. This signal does not record litigation history of any other kind.
Bar Guidance Alignment
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. No bar opinion is named, ABA Formal Opinion 512 does not appear, no state guidance on generative AI in legal practice is referenced, and nothing maps the product to any jurisdiction's rules of professional conduct. Nor is professional responsibility referred to in general terms, which is what separates this from the records that take the tier above: neither published document mentions professional obligations, competence, supervision or the practice of law at all.
The Terms of Service, read in full, are a website terms of use running to twenty-three sections and containing no reference to legal advice or to the professional context in which the product is used; the Privacy Policy is a data document that excludes enterprise customer processing from its scope. The absence sits against a product that recommends whether to settle or fight a live matter and is marketed to attorneys, paralegals and legal operations staff.
Recorded as an absence in the published record on the date shown, not as a finding about the vendor's internal practice.
Searched the product overview, security page, Trust Center, Terms of Use, privacy policy and newsroom on 18 September 2026. No ethics opinion, court AI order or rule, or bar guidance is referred to.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Time savings are claimed and nothing addresses billing or disclosure, on a product that sits inside a lawyer-client fee relationship for at least part of its published market. The savings claims are the headline of the vendor's own announcements: that the average mid-sized firm reviews roughly 650 cases a year taking between seven and thirty days each, and that with this product the time is compressed into seconds; and on the current site, that legal teams resolve matters faster and do more with the same headcount.
The fee relationship is engaged because law firms are a named buyer in the vendor's own material, the May 2025 announcement describing settlement prediction tools tailored for Big Law firms alongside general counsel, and case assessment is conventionally billable work. Nothing published addresses what happens to the bill when an assessment that took days takes seconds, whether AI-assisted analysis is identified on an invoice, or whether a client is told the tool was used.
Recorded because it complicates the picture rather than changes the value: the product's current positioning is toward in-house departments and, earlier, litigation funders, and for those buyers no client bill is in the loop.
Billing economics are marketed without disclosure guidance. The vendor says firms report 60 percent fewer non-billable hours per matter, 10 percent fewer write-offs and profitability on alternative fee arrangements, and generates carrier-ready status reports; nothing addresses how firms should bill or disclose AI-assisted work to clients or carriers.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
No located public material supports a client-side disclosure obligation, and none of the three artifacts this signal looks for exists. There is no subprocessor list on any surface. No model provider is named anywhere, so a firm or a company asked by its counterparty which third party sees its case files cannot answer from anything this vendor publishes; the only provider information in circulation is third-party press reporting and is not the vendor's disclosure.
There is no client-facing disclosure material, no data processing addendum, no security exhibit and no consent or notification pack that could be forwarded. The value is not on-request either, because nothing indicates the material exists behind a sales conversation: the estate contains no security or compliance contact route, no trust center and no request mechanism of any kind, so under the discipline that a value's own words must be true, on-request would assert the existence of something no surface evidences.
What is published instead is a website Terms of Service and a Privacy Policy that expressly excludes enterprise customer processing and routes it to unpublished customer agreements.
Disclosure material is available on request. The privacy policy says business customers can get the subprocessor list from their account representative, the security page says Turbo Law answers security questionnaires as part of deployment and signs BAAs, and the Trust Center offers more than 150 documents on request. No list is published.
Court Disclosure Support
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. One element sits close enough to be worth recording and is deliberately not credited: the vendor states that every case is summarized with clear attribution to source material, so a reader of a summary is told where a proposition came from within the customer's own file. That is a usability and trust claim addressed to the person reading the summary, not a record addressed to a court or a certifying lawyer, and it is a single marketing sentence with nothing behind it.
Nothing states that any output identifies which model produced it, and no model is named anywhere in any event; nothing distinguishes machine-generated content from human-edited content in an exportable form; no audit trail, export or per-document record is described; and no disclosure template, certification form or court-facing guidance was located. The question has real weight on this product, because a settlement recommendation or exposure figure informs a decision a party may later have to explain, and because at least one published matter category, regulatory and compliance matters, sits close to proceedings where the basis of an assessment can be probed.
Parts of a record exist. Every access, query and export is logged and exportable, and outputs carry line citations to their sources; no record of which model produced an output or who verified it is described.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Commercial Transparency
- Primary Law Corpus Provenance
- Good Law Verification
- Bar Guidance Alignment
Which one fits
Choose Theo Ai if
- You need to see your whole litigation portfolio at once. Theo Ai collects case material automatically from Microsoft, Google, Workday and Salesforce, turns emails, invoices and filings into structured data, and shows every matter's cost trajectory and risk signals in one place, with each case summarized and attributed to its source material.
- You need to decide which claims to settle and which to fight across a volume of matters. Theo Ai's prediction engine recommends settling or fighting each claim, tracks exposure as new facts arrive, and flags matters it judges to be under reserved.
- Your caseload is general liability, workers' compensation, employment, regulatory, vendor disputes or internal investigations in retail, insurance, chemicals, hospitality or life sciences. Theo Ai publishes indicative matter counts for each of those six categories and names those five sectors as its market. Onboarding is currently by invitation, with a public waitlist.
Choose Turbo Law if
- You need drafts and analysis built from one matter's whole file. Turbo Law reads the file into a fact graph of parties, experts, treaters, admissions, defenses and dates, each line cited, and produces chronologies, deposition contradiction reports, privilege flags, motions and discovery responses from it.
- You need a case evaluation that stays with the attorney. Turbo Law's evaluations, settlement value models and carrier status reports surface for attorney review before anything goes external, and its stated rule is that a fact it cannot cite to the file or the public record is not claimed.
- Your security reviewer needs written commitments before files move. Turbo Law publishes that customer content does not train foundation models, that it neither asserts nor waives privilege, that each firm runs in an isolated Azure environment, and that SOC 2 Type II and HIPAA sit in its trust center.
In summary
Theo Ai
Theo Ai, from Theo AI, Inc. in Palo Alto, sells predictive litigation intelligence to defense side corporate legal departments: a portfolio command center that collects case material from Microsoft, Google, Workday and Salesforce, presents every matter's cost and risk in one view, flags under reserved matters and recommends which claims to settle and which to fight. It began in 2024 selling case outcome prediction to litigation funders. The AI Legal Index grades it in the top two bands on two of fifteen capability axes, with an A on AI centrality and a B on practice coverage, where it names six matter categories and five sectors. As of 12 September 2026 the index located no customer agreement, no training position, no security attestation and no oversight structure for its recommendations.
Turbo Law
Turbo Law, headquartered in San Mateo, California, sells an AI litigation platform to defense law firms and insurance carriers for complex matters including medical malpractice, mass and toxic tort, transportation, employment and construction. It reads a whole matter file into a line cited fact graph and generates chronologies, contradiction reports, motions, discovery responses, case evaluations and settlement value models, with an assistant for multi step tasks. The AI Legal Index grades it in the top two bands on ten of fifteen capability axes. It states that high stakes outputs go to attorney review, that customer content does not train foundation models, and that it runs in dedicated Azure environments under SOC 2 Type II. As of 18 September 2026 the index located no named customer, no published price and no platform agreement.
Questions buyers ask
Are Theo Ai and Turbo Law the same kind of product?
No. Theo Ai works at the level of a whole litigation portfolio, showing an in house legal team the cost and risk of every matter and recommending which to settle and which to fight. Turbo Law works inside one matter, turning its file into chronologies, motions, discovery responses and a case evaluation for defense counsel. Both sell to the defense side, and both estimate what a case is worth, which is why the AI Legal Index grades them on the same fifteen axes.
Does Theo Ai train on client case files?
Theo Ai publishes no position either way. Its terms govern its website, and its privacy policy states that it does not apply to information processed on behalf of enterprise customers, which is governed by customer agreements that are not published. A company sending case files to Theo Ai would need that term in its own agreement. Turbo Law states that customer content does not train foundation models, a commitment it says its contracts enforce. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
What does Turbo Law publish about security?
Turbo Law's trust center lists SOC 2 Type II and HIPAA as compliant, names Prescient Security, and offers more than 150 documents on request, with SOC 2 documentation available under NDA. Its security page describes dedicated Azure environments with tenant isolation, encryption at rest and in transit, enforced two factor authentication and exportable logs of every access, query and export. Theo Ai publishes no certification or security page. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
How accurate are Theo Ai's settlement predictions?
Theo Ai publishes no accuracy figure, test set or evaluation for its predictions. Its only published grounding claim is that every case summary carries attribution to source material. A figure of about 85 percent predictive accuracy appears in a December 2025 press interview with its chief executive, with no test set or method, and is not a vendor disclosure. Turbo Law says its workflows are tested against accuracy standards before release but publishes no results. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
What do Theo Ai and Turbo Law both leave unpublished?
Who answers for a wrong valuation. Both produce estimates of what a case is worth, and neither publishes the customer agreement that would allocate the loss if one is wrong, nor any measured accuracy for those estimates. Neither names the models it uses, publishes a price, or engages with bar guidance on AI. Neither says how a lawyer relying on a recommendation should disclose it, or in which region its data is processed. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
This page compares two adjacent products rather than two answers to one purchase, and the grid is its purpose. Theo Ai's low grades record that the documents it publishes do not reach its product: its terms govern its website, its privacy policy excludes the enterprise data it processes, and no customer agreement is published. A figure of 85 percent predictive accuracy attributed to Theo Ai's chief executive appears in press coverage and is not the vendor's own disclosure. Theo Ai was verified on 12 September 2026 and Turbo Law on 18 September 2026. Neither vendor reviewed this page.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything above comes from public material on the dates shown. How the index grades.