S

Skribe

The models run the working layer, on a product whose official output is checked by people. Speech recognition produces the rough transcript and powers the Library product, and generative AI finds key moments, summarizes testimony and answers questions across a deposition with links back to the video. The official record, though, is the recording itself, and a person proofreads and swears to the final transcript. The hosted tier also includes a digital reporter and recording management that work without the models. Remove the AI and a customer still has a recorded deposition transcribed by a human.

Answers are grounded in the record with a direct link to source, and one accuracy figure is published without a method. The AI page says every AI answer is supported by links to the exact point in the video, so any answer can be checked against the testimony. The home page states that final transcripts proofread by people are 99 percent accurate and compares that with the 95 percent threshold for stenographer certification.

That figure covers the verified final transcript only, not the rough transcript or the AI summaries and answers, and no method, sample or test is described. The accuracy of the rough transcript is not stated.

Human verification is built into the official deliverable, and the AI analysis layer has no stated review. The vendor separates the same day rough transcript, produced by speech recognition, from the final transcript, which it says is proofread by a person every time and comes with a sworn Declaration of Verbatim Transcript. It states that the audio and video recording is the official record. A trained staff member attends hosted depositions and monitors the recording.

The AI summaries, key moments and answers carry no equivalent control, and nothing states what the rough transcript should not be used for.

Named customers are published, without figures. The home page carries attributed testimonials from Charles Peckham of Peckham Martin, Steven Sachs of Espinoza & Brock and Randy Sorrels, managing partner of Sorrels Law, alongside firm logos. The testimonials describe faster transcripts, video the same day and lower overhead, but give no measured result. The vendor's cost comparison with traditional court reporting is a pricing claim, not a customer outcome.

Confidentiality for uploaded material is disclaimed in the published terms. Section 13 of the Terms & Conditions, revised 27 July 2026, states that Skribe does not guarantee any confidentiality with respect to submissions, alongside a broad perpetual license over uploaded content. The section 12 confidentiality clause does not cover testimony the service uses for its intended purposes. Section 4.c, added in July 2026, keeps a backup recording running while the meeting room is open, including recesses and breaks, and attendees are advised to mute or leave the room. Nothing addresses privilege or work product.

A real position is published on what the service is and where it lawfully operates, and supervision is not covered. The terms state that Skribe is not a law firm and gives no guarantee that recorded testimony is admissible. The site states that Skribe digital reporters are not certified shorthand reporters and do not perform stenography, and that depositions are conducted under rules permitting recording without a stenographer, with those states listed on the admissibility page.

Subscribers are defined as attorneys and law firms. Nothing addresses an attorney's duties when relying on AI summaries or answers, and the rough transcript is not set apart from the sworn final in professional responsibility terms.

No governance position is published. There is no responsible AI policy, testing regime or accountable owner, and nothing discloses how speech recognition performs across accents, speakers or audio conditions. That gap matters for a product whose transcripts become evidence. The site's main pages, the terms and the privacy policy were searched.

Retention and security are covered only in general terms. Personal information is kept indefinitely by default, and after cancellation Skribe has no obligation to preserve data. The only security measure named is SSL encryption, and internal access is restricted to staff with an operational need. No encryption at rest, subprocessor list for recordings or AI processing, or incident notification commitment is published.

Limitations in the terms disclaim the exposure the product creates. The software is provided as is with no warranty of accuracy or completeness, and Skribe's liability is capped at the amount paid and never more than $1,000. A user who pays nothing is limited to injunctive relief. Users must indemnify Skribe, including for reliance on advice provided through the software. No vendor indemnity or insurance position is published.

No integration into practice systems is published. Skribe is built on Zoom, and its Skribe Link feature records Zoom depositions a firm does not host; both are capture channels, not practice systems. No connection to document management, case management, discovery or trial presentation software is named, and clips are shared by link. The home, product and pricing pages and the terms were searched.

Location is stated, and hosting and tenancy are not. The terms say the software is operated in the United States, and the privacy policy says personal information is stored and processed on servers there. No hosting provider, tenancy model or deployment option is named, and where speech recognition and AI providers process recordings is not published.

No independent security attestation is published. The home page, terms, privacy policy and FAQ were searched, no security page is linked from the navigation, and a targeted search for a Skribe SOC 2 or other attestation found nothing. The only security detail published is SSL encryption at 128 bit strength, in the privacy policy.

The models underneath are not identified. The vendor refers to its automated speech recognition system and generative AI tools but names no speech recognition provider, language model or model provider. The search covered the home page, AI page, product pages, terms and privacy policy.

A deposition can be priced before speaking to anyone. Hourly rates for both products, what the hosted tier includes, the surcharge for sessions held in person, and the cancellation and billing terms are all published on the home and pricing pages, and the pricing structure is recorded in full. What volume or firmwide terms look like is not stated.

The working boundary is published state by state, with its conditions. The admissibility page tables 42 U.S. states plus the District of Columbia and the federal rules, marking each as allowed, allowed with stipulations, or allowed with stipulations to remote proceedings. Each entry cites the governing rule, for example FRCP 30(b)(3), Cal. Code Civ. Proc. section 2025.330 or Tex. R. Civ. P. 203, and offers a sample notice for recording without a stenographer.

The eight states not listed, among them New Jersey, Connecticut and Washington, fall outside the published boundary. The vendor serves litigation teams at firms of every size and covers depositions, examinations under oath and other sworn proceedings, remote or in person. Nothing states which proceeding types or languages the speech recognition and AI handle less well.