L

LegalSifter

ReviewPro uses models to interpret contract language, detect playbook issues, propose edits and answer questions through a contract assistant. The playbook provides the organization's positions and fallback instructions, so interpretation and generation are essential to the review service. The current product also includes a versioned contract repository, intake ticketing and controlled playbook management. Those organizational functions still have a use when an AI review is not running.

LegalSifter's human services can customize playbooks or handle redlining and extraction, while Contract Logix is a separate contract management offering within the company. The broader workflow now extends beyond the original automated review engine.

ReviewPro 2.0 evaluates clauses against a selected playbook and returns proposed edits with rationale for the reviewer. Its July 2026 study used 30 NDAs, 16 positions and three runs per agreement, comparing results with attorney approved edits. With the engine and agreements fixed, richer playbook instructions raised issue detection from 91% to 95%, selection of the right action from 77% to 93%, and edit fidelity from 55% to 93%.

These are distinct measures of contract review, not a general guarantee of legal accuracy. The benchmark does not cover citation treatment, an abstention threshold or performance across other agreement types.

ReviewPro runs a first pass against the selected playbook and returns redlines and explanations in Word or Google Docs. The customer must vet standard or custom playbooks before using them, and reviewers remain responsible for checking and accepting the result. The terms impose an additional categorical boundary on Trial Features: they are solely for evaluation and testing, and cannot be used in production. Named seat plans also require interactive human use rather than unattended batch processing.

Routine business review can follow the legal team's approved standards, but the terms recommend legal counsel and leave responsibility for the final agreement with the customer.

A July 2026 study describes a field observation involving 15 complex agreements, where an unnamed legal team reached 90% alignment with its intended edits and reduced a roughly 90 minute review to about five minutes. That account is separate from the controlled 30 NDA benchmark used to test playbook instruction quality. The field result indicates a concrete workload and reported change, but gives neither the customer's name nor a detailed timing protocol.

The internal benchmark explains its agreements, positions and repeat runs; it does not turn the field observation into a reproducible customer deployment study across different contract portfolios.

The ReviewPro agreement treats customer contract content as confidential and expressly intends disclosure to preserve attorney client privilege and work product protection. Access by service providers is limited to those who need the information and owe confidentiality duties. Tenant records are isolated from other customers, with role and feature controls inside the service. Training third party AI models is prohibited, but the terms grant LegalSifter a perpetual right to train and improve its own models with anonymized customer content.

The security FAQ describes training as a choice made in the agreement. This default permission and the absence of a stated internal matter wall limit the confidentiality position.

LegalSifter identifies ReviewPro as software, disclaims legal advice and an attorney client relationship, and strongly encourages review by legal counsel. Customers must vet their playbooks and verify the resulting work. Business personnel can use standards approved by their legal team, while the agreement leaves legal decisions with the customer. Subscription scope also matters: named seat plans are restricted to internal contracts and matters, while legal services for external clients require express authorization under a consumption plan.

The product supports supervision through explained redlines and reusable standards, but does not set jurisdiction specific practice limits or determine whether a particular user may give advice.

LegalSifter separates issue detection, selection of the appropriate action and edit fidelity when evaluating ReviewPro. Its July 2026 study holds the review engine and NDA set constant while varying the specificity of playbook instructions, with attorney approved edits as the comparison. It also compares releases and describes recurring assessment of thoroughness, accuracy and readability. These mechanisms connect product changes to defined review tasks rather than a general responsible AI promise.

The terms prohibit discriminatory consequential decisions and malicious attempts to manipulate models. LegalSifter names no accountable AI governance owner and gives no results on uneven behavior across populations or contract categories.

ReviewPro uses encrypted contract storage, AWS key management, role and feature controls, and tenant isolation at the record level. Auth0 supports SAML and OIDC single sign on, with the customer's identity provider enforcing multifactor policies. Contracts are deleted within 30 days after subscription termination, while no ordinary retention period is stated for the full prompt and output history. OpenAI and Anthropic are listed as model processors in the United States, and the terms prohibit their training on customer content.

Confirmed customer data or model integrity incidents trigger notice without undue delay, including incidents involving poisoning or model extraction. Retained anonymized material has separate contractual treatment.

The ReviewPro terms provide a vendor defense and indemnity for specified claims against the authorized service, but expressly exclude AI output, reviewed contracts and customer playbooks from that protection. LegalSifter disclaims output accuracy and liability for decisions based on the service. The general liability cap is the fees paid during the preceding twelve months, with exceptions including indemnification obligations, gross misconduct and applicable legal duties.

Trial Features carry separate warranty and indemnity exclusions and cannot be used in production. These terms allocate platform infringement risk more specifically than a generic disclaimer, while leaving errors in an accepted redline and resulting legal decisions with the customer.

ReviewPro runs as a Microsoft Word add in applying tracked redlines in the document, and as a Google Docs add in whose marketplace listing documents exactly what it does: reads the current document to extract contract text, creates a redlined copy in Drive, transmits content to LegalSifter's servers and stores nothing locally, with the Google permission scopes listed. The product page states connection to existing apps and CLM software without naming them beyond the company's own Contract Logix.

ReviewPro operates on AWS with multiple availability zones, encrypted storage and customer separation enforced at the record level. OpenAI and Anthropic appear in the model processor register with United States locations. The security description covers resilience, identity and tenant controls, but does not name an available application hosting region or a customer residency selector. A model provider's listed country is not a complete account of where every processing step takes place.

Contract Logix belongs to the same company but is a separate platform, so its Azure hosting arrangement has no bearing on ReviewPro's architecture or a deployment option for this review service.

LegalSifter states that ReviewPro's security controls undergo independent SOC 2 Type II auditing. Its security material describes AES 256 encryption, TLS 1.2 or later, and AWS key management using FIPS 140 2 Level 3 validated hardware. Staff background checks, security training, monitoring and identity controls accompany the attestation claim. The certification description does not identify an audit coverage period, named assessor or freely accessible report with a scope statement.

Encryption standards and a cloud provider's validated hardware support the operating controls but are separate from the scope of LegalSifter's own independent audit.

The current processor register names OpenAI and Anthropic as large language model and API suppliers for ReviewPro, both in the United States. The security explanation discusses OpenAI's business terms, while the service agreement requires third party AI components not to train on customer content and to follow stated retention controls. LegalSifter separately reserves the right to develop and improve its own models from anonymized customer material.

LegalSifter gives no exact model versions, no mapping of review tasks to each provider and no commitment to notify customers of every model substitution. Earlier marketplace descriptions of provider choices do not fully describe this current supplier list.

ReviewPro offers annual volume subscriptions with unlimited seats or annual seat subscriptions with unlimited interactive contract reviews. Volume plans cost $2,750 for 50 reviews, $7,500 for 150, $18,000 for 400 and $40,000 for 1,000. Seat pricing starts at $6,000 a year for one user; larger minimum seat commitments reduce the rate per user. Standard implementation is included at no extra charge. Custom playbook architecture and ongoing custom maintenance are separately scoped optional services.

Named seat subscriptions are limited to internal contracts, while legal client work requires express permission under a consumption subscription. The plan choice therefore changes both the charging unit and permitted use.

ReviewPro supports legal departments and business teams handling recurring commercial agreements against approved positions. Standard playbooks cover common forms, and custom architecture lets a legal team encode its own preferred and fallback language. Industry offerings include healthcare, insurance, manufacturing, technology, higher education, staffing and law firms. A firm's plan matters because named seat subscriptions cover only its internal contracts, not an unlimited external client review service.

Client work requires express authorization under a consumption subscription. The terms also restrict protected health information to an executed BAA and eligible features. The full set of supported jurisdictions and excluded agreement types is not defined.