V

Verbit

The models are the engine of the product, and the business survives without them. Captivate is the vendor's own automatic speech recognition engine, trained on legal terminology, adapting to customer vocabulary, speakers and acoustic conditions, and Gen.V is a generative layer producing summaries, chapters, keyword tagging and inconsistency detection. Every legal plan is priced and described around that engine. Verbit also sells the same output produced by people: human review, certified transcripts and the VITAC, Take Note and AST brands are human transcription operations.

The Terms describe the Services as processing live and prerecorded media using ASR or human options. Remove the models and a transcription business remains.

The product cites no legal authority, so authority grounding and citator checks do not apply; grounding and measured accuracy do. Grounding is strong: output is time coded and labeled by speaker against the source recording, so every word can be traced to audio a reader can replay, and a published Post-Production Corrections Policy covers fixes. The hallucination position is in the contract. Section 4(d) of the Terms says AI Analysis Features use probabilistic models that may produce inaccurate, incomplete or misleading results and are not professional advice.

It adds that they are no substitute for qualified human review and that the customer must verify output before relying on it. No accuracy figure, test set or error rate is published, and the accuracy commitments in section 7(f) are targets that depend on reasonable audio quality as judged by the vendor.

Modes, the constraint, the review surfaces and the route back to human judgment are all published. The modes are the product tiers: AI only transcription, AI with optional human review, and certified transcripts completed by people, each described and priced separately. The constraint is printed on the plan: the self serve Standard legal plan lists Not court admissible among its features, so a buyer knows before buying what unsupervised output may not be used for.

Court admissible transcripts are available on request on higher tiers, and each names an upgrade path to certified transcripts. The review surfaces are the editing workspace, rough drafts set against final transcripts, and the Post-Production Corrections Policy in the Terms. Section 4(d) of the Terms puts verification of all AI output on the customer and says the features are no substitute for qualified human review.

On a transcription product the boundary is a type of document rather than a confidence score, and the vendor draws it plainly.

Figures are published without named customers. The vendor says it serves more than 3,000 businesses and institutions, runs the VITAC, Take Note and AST brands, and sets a 99.0% monthly availability target in section 7(b) of the Terms with defined exclusions, which is a measurable commitment. Turnaround times are committed in the Order or on the site and counted in US Eastern business days. No named law firm, court or court reporting agency appears with a figure for what changed, and no dated case study for the legal line is published. Analyst rankings and awards are not evidence of deployment.

Privilege is addressed in the contract. Section 5(c) of the Terms, headed Privileged Materials, applies where a customer submits files containing attorney client privileged or work product material. It commits the vendor to limit access to staff who need it to provide the Services, and to disclose to no third party except as needed to provide the Services and then under equivalent confidentiality. The vendor may use the material only to provide the Services and for customer specific optimization, and must return or destroy it on termination or request.

Nothing in the agreement waives any privilege held by the customer or its clients, and the customer decides whether submitting is consistent with professional conduct rules and privilege law. Work product is named expressly. Section 4(c) bars generative AI training on customer content, and section 4(e) bars general speech recognition training on law firm material without written opt in. DPA section 8 requires deletion or return within 30 days of termination.

Access control is documented on the Trust Page. Third party model providers may be used only if they are barred from training on customer data and keep data no longer than their own terms require.

A real position on the advice line is published, without the supervision and competence side. Section 4(d) of the Terms says AI Analysis Features are not professional advice of any kind, are no substitute for qualified human review, and must be verified by the customer before reliance. Section 5(c) goes further, naming the customer's professional conduct rules and privilege law as the customer's responsibility to assess.

For this kind of product the advice line is the admissibility line, and the vendor draws it: the AI only plan is marked not court admissible, certified transcripts are an upgrade, and jurisdictional formatting and cover pages are what make a transcript admissible. Nothing addresses a supervising lawyer's competence duties when relying on machine transcription of testimony, and nothing says who may use the service. Nothing addresses state licensing for digital court reporting, the live professional question for this kind of product.

A governance framework with real substance is published, without testing results or a named owner. It sits in the contract rather than on a principles page. Section 4 of the Terms covers customer specific optimization, general speech recognition training with requirements to remove identifying details, a ban on generative AI training, generative AI features, legal provisions and FERPA. Section 4(g) commits the vendor to technical and organizational measures that prevent reidentification of training data, restrict access to identifiable data, and keep data in AI Analysis Features encrypted in transit and not retained by third party providers.

The DPA commits that automated processing will not produce legal effects on, or significantly affect, data subjects without safeguards. Security governance is documented separately: a security management system led by a named CISO role, risk assessments reviewed by senior leadership at least yearly, annual policy reviews, and annual and quarterly internal audits. No accountable owner for AI is named, and no evaluation results or model card are published.

Nothing addresses uneven output across accents, dialects or speaker groups, the bias question that matters for transcribing testimony.

Most of the ground is contractual and specific; the gap is the subprocessor list. Section 8 of the DPA requires deletion or return of all personal data within 30 days of termination at the customer's choice, keeping one copy only for legal claims, and the Trust Page says a customer can have data removed at any time through its account manager. Access control is documented in detail: HTTPS enforced in transit, AES-256 encryption at rest in AWS S3, two factor authentication, roles the customer assigns, least privilege internally, regular access reviews and rules for staff joining, moving and leaving.

Incident notice is contractual: notice without undue delay after the vendor learns of a data incident, with enough information for the customer to meet its own notice duties. A CISO led response team and a 24/7 security operations center back it. No subprocessor list is published. The DPA commits to notify before authorizing a new subprocessor, to provide the current list on request, and to a five business day objection window, but a buyer cannot see who processes its recordings before signing.

Specific caps and a real indemnity are published, short of the exposure the product creates. Section 8 of the Terms caps each party's liability at the lesser of USD 1,000,000 or the fees paid or accrued in the 12 months before the claim, excludes indirect and consequential damages, and carves intellectual property violations and the customer's payment duties out of the cap. Section 6(b) gives a three part indemnity from the vendor.

It covers third party IP infringement from authorized use, and unauthorized disclosure of confidential information or personal data caused by the vendor's negligence or failure to follow its published Security Standards. It also covers violation of law caused by the vendor's provision of the Services. The second part is a real data breach indemnity, measured against the Trust Page. Availability is committed at 99.0% a month, and a Post-Production Corrections Policy covers defective output.

Nothing allocates the loss when a transcript of testimony is wrong: section 4(d) puts verification of all AI output on the customer, and section 7(g) disclaims all warranties, including fitness for purpose. No insurance position is published.

Real integrations are documented, without depth across the legal stack. A public API with developer documentation is published at verbit.readme.io, and API access is included in the Custom plans. Named integrations include Relativity on the Legal Visor plan, which reaches the ediscovery platform where litigation work already sits, plus Zoom and Teams for live capture, cloud storage destinations and a mobile app for recording.

The legal transcription material says the API can connect transcripts directly to case management and document systems, but none of those systems is named. The integrations page was not opened and may say more. The repeated phrase about dozens of video, conferencing and cloud storage integrations describes breadth without a list an implementer could use.

Real residency detail is published, without the full picture. The Trust Page says delivery is on AWS, replicated across several availability zones and regions under the AWS uptime commitment. The privacy policy names processing locations: the United States, Europe, the United Kingdom, Israel and elsewhere as needed. It says customer data may be processed only where the DPA and other agreements allow, so residency is a term that can be negotiated.

Schedule 2 of the DPA handles transfers with the EU Standard Contractual Clauses, Modules Two and Three under Irish law, the UK Addendum and Swiss adjustments, plus extra safeguards including a commitment to resist bulk surveillance requests. The tenancy model is never stated, no region menu or single region option is published, and the split between where data is processed and where it is stored is not described.

Certifications are real, named with versions and scoped, and the evidence is gated. The Trust Page says ISO 27001:2022 certification and SOC 2 attestation are available for certain services and configurations and directs buyers to an account manager or support for details; that scoping caveat is itself a disclosure. The security program around it is published in detail. It includes monitoring through GuardDuty, CloudWatch and CloudTrail, cloud security posture management with scanning of infrastructure code, a SIEM, and a CISO led incident response team with a 24/7 external security operations center.

Vulnerability fixes are targeted at 48 hours for critical and 14 days for high severity, development follows a documented secure lifecycle with code review and OWASP Top 10 alignment, and an independent penetration test is commissioned at least yearly. The DPA gives an audit right on 14 days' notice once every 12 months, including by the customer's own third party auditor. The certificate and report are available only through a sales or support conversation. No auditor, report period or certificate number is published, and there is no self serve trust portal.

The vendor's own engine is named, and what sits under the generative layer is not. Captivate is named as the proprietary speech recognition engine, and its behavior is described, including adaptation to legal language and to each customer. Beyond that, the data policy says only that Verbit integrates large language models to improve transcription and analysis, and Gen.V is named as the generative suite, but no model or provider is identified anywhere.

The position is contractual. Section 4(c) of the Terms allows only third party AI providers that are barred from training on customer data. The DPA requires any such provider to be listed as a subprocessor, with notice before it starts processing and a five business day objection window. A customer can learn who the providers are and object, but only by asking, since the subprocessor list is not published.

Real pricing is published for part of the range, with the structure published for all of it. The legal Standard plan costs $24 a month for 100 hours, 50 live and 50 post production, with self serve registration rather than a sales call. The plan page sets four legal tiers side by side, comparing hours, languages, integrations, user counts, security features, admissibility and upgrade paths, and the same structure is published for media, corporate and education.

Commercial terms sit in the Terms. They set payment in advance unless an Order says otherwise, 30 days' written notice of any price change taking effect at the next renewal, a 60 day billing dispute window, and interest on overdue amounts at the lower of 1.5% a month or the legal maximum. They also allow suspension on 10 days' notice for nonpayment and termination for convenience on 60 days' notice. Three of the four legal tiers, including both products an attorney or agency would actually buy, have no price and lead to a demo.

Who the product serves is documented precisely, and the limits are stated. Dedicated pages cover law firms, courtrooms, law enforcement and government, alongside corporate, education and media. Within the legal line, each buyer type has its own product: Legal Visor for attorneys and litigation teams, Legal Capture for court reporting agencies and digital reporters, courtroom material for courts and agencies, and separate material for legal departments.

Supported proceedings are listed: depositions, trials, bench hearings, jury trials, arbitrations, witness interviews, dictation, and digital evidence including body camera footage and 911 calls. The limits are published at the point of sale. The entry plan is marked not court admissible, limited to one user and to English transcription, with basic export and integrations. Higher tiers state where Spanish is added, where certified transcripts are an upgrade, and where court admissible output is available on request. The unit of work is a proceeding, not a practice area, so practice area coverage does not apply.