O

Opus 2

The models drive several capabilities a buyer would miss, on a platform that ran without them for most of its life. Opus 2 Ingenuity is presented as an agentic framework covering document summaries and analysis, extraction of people, organizations, events and legal topics, questions scoped to a matter or selected documents, saved prompts, drafting and checking work product, speech recognition for legal transcripts, and live hearing analysis that compares testimony with the bundle to surface inconsistencies, which depends most on the machine.

Underneath sit eighteen years of case management, shared workspace, annotation, chronologies, electronic bundles, evidence presentation and hearing infrastructure, none of which needs a model. The vendor places the AI on top of that platform, describing the platform as the source of the matter intelligence the AI uses.

Quality is asserted repeatedly and never measured. The vendor says better data relationships give more accurate insights, AI features add only relevant and helpful information, and the design stops errors from harming a case. Answers are said to be based on the selected documents and surrounding matter data, with the system locked to a single matter. Checked the AI, home, security and arbitral institutions pages: no accuracy figure, test set, benchmark, evaluation, error rate or hallucination disclosure.

Nothing says whether an answer or summary cites its source document, or whether a reader can open the source from the output. This matters most in hearings, where the system flags inconsistencies between live testimony and the bundle and counsel may act within minutes. Citator checks do not apply to a platform working on a matter's own evidence.

A clear limit on autonomy is published, with a worked example. The Control principle says lawyers must control the process, using their own knowledge and judgment, before any data is saved permanently, so AI output stays a suggestion until a person commits it. The chronology feature shows this: the vendor says auto generating a chronology fills a timeline with irrelevant placeholders a team must remove, so instead the AI surfaces events for the lawyer to assemble, deciding relevance themselves.

Entity extraction works the same way, showing people, events and organizations with summaries so the user can drop irrelevant ones and duplicates before anything is added. Checked the AI, home and security pages: no threshold at which a feature declines to answer, no confidence signal, no route to report or fix a wrong output, and nothing on what happens after an error reaches a work product.

Deployment evidence is deep, named and current, with no measured outcome. Five customer quotes give name, title and firm: a contentious legal technology portfolio manager at Linklaters, the chief knowledge and innovation officer at Dorsey & Whitney, the director of practice technology at Vorys, a litigation partner at Pillsbury and a paralegal at Shepherd and Wedderburn, plus two chambers clerks. About nineteen firm logos appear, including Baker Botts, Dentons, Herbert Smith Freehills Kramer, Holland & Knight, Mishcon de Reya, O'Melveny, Troutman Pepper Locke and Wilson Sonsini.

The institutional deployments can be checked independently: SIAC Gateway, run by the Singapore arbitral institution SIAC, is said to run on Opus 2, the Dubai institution DIAC is a named partner, and the transcript and hearing services supported the Manchester Arena Inquiry, Grenfell Tower Inquiry phase 2 and the Hillsborough Inquiry. The vendor reports use by 98 percent of Am Law 50 firms and 92 percent of the Financial Times top 25 most innovative firms. Those figures show reach, not results; no quote carries a number or date, and no case study with a method was found.

One precise commitment covers the question that matters most, and nothing else. The Secure principle says client data will never be shared outside the client's instance or between two or more cases inside it, an explicit undertaking across matters and tenants, reinforced by queries locked to a single matter or selected documents and by firewalls, encrypted volumes, logical identifiers and ownership tags. The arbitral institutions material acknowledges that arbitration is confidential.

No customer agreement is published, so none of this is contractual: the footer carries a privacy policy, cookie policy, modern slavery statement, supplier code and code of business conduct, but no terms of service, master agreement or data processing addendum. Nothing addresses training on client material, no retention position exists for documents, prompts or output, and privilege and work product are not addressed in writing, though the platform holds working papers from highly sensitive disputes and public inquiries. The privacy policy was not opened and may say more.

Nothing published addresses the advice line, and no document on the site would carry it. The footer, read in full, lists a privacy policy, cookie policy, modern slavery and human trafficking statement, supplier code of conduct and code of business conduct, with no terms of service, disclaimer or professional responsibility statement. Nothing says output is not legal advice, nothing addresses responsibility when a drafted document or flagged inconsistency proves wrong, and no bar or regulator is named, including ABA Formal Opinion 512 and, for a London vendor selling to English firms and chambers, any Solicitors Regulation Authority or Bar Council guidance.

What is published is a supervision stance rather than a disclaimer: the vendor says it is not trying to replace a lawyer's intelligence, the AI enhances human ingenuity, and lawyers must apply their own judgment before anything is saved. That describes how the work is done, not who answers for it.

Five well written principles are published, but not a governance framework. Empower, Enhance, Control, Secure and Trust each get a paragraph explaining design choices and rejected alternatives: why the vendor does not auto generate chronologies, why it keeps existing legal workflows, and what the AI may not write into a case without a lawyer's action. Two carry real commitments, on human control before anything is saved and on data never crossing an instance or case boundary.

Searched the AI, home, security and arbitral institutions pages and found no accountable owner for model behavior, no testing regime for the AI before release, no published results, no model documentation, and no standard or certification for AI as distinct from information security. STRIDE threat modeling and CREST accredited penetration testing cover software security, not model behavior. Nothing addresses bias or uneven output, though the hearing feature analyzes live testimony, and how it performs across accents, languages and interpreted evidence in international arbitration goes unasked.

Strong on how the system is built, and silent on what happens to the data inside it. Published: AES-256 encryption at rest, TLS 1.2 in transit, including between internal components, email encrypted end to end, logical data separation through firewalls, encrypted volumes, logical identifiers and ownership tags, client defined password policy, several multifactor options, single sign on through the client's provider, and full client control over users.

Development practice is well documented: STRIDE threat modeling with a risk owner per threat, regular CREST accredited external penetration tests, automated and manual code review including third party dependencies, continuous integration, and formal change management with risk and impact assessment. Searched the security and AI pages and the footer and found no retention period, deletion or return commitment, export provision, subprocessor list, named cloud provider, or incident response or breach notification practice. The privacy policy was not opened and may say more.

Nothing is published on who bears the loss when the system is wrong, because no customer agreement is published. The footer legal inventory, read in full, is a privacy policy, a cookie policy, a modern slavery and human trafficking statement, a supplier code of conduct and a code of business conduct in PDF. There is no terms of service, no master services agreement, no subscription terms, no service level agreement and no data processing addendum.

Nothing therefore states a liability cap, an indemnity in either direction, a warranty on the software or on generated output, an insurance position, a limitation period, a governing law or a forum. The security page refers in passing to Opus 2 Services and Agreements meeting global legal and regulatory requirements, and names six regimes they are said to satisfy, which confirms agreements exist without publishing any of them. A buyer cannot read the allocation of risk in advance at any level.

Integration is mentioned by category, never by name. The AI page says core documents flow from ediscovery and document management systems into one workspace, naming neither. The security page describes linking to a client's own single sign on provider, a useful identity integration but not a practice system one. Checked the AI, home, security and arbitral institutions pages and the full navigation: no document management system such as iManage or NetDocuments is named, and nothing covers matter management, ebilling, court filing, Word or Outlook.

No API, developer documentation or integration list was found. The vendor runs hearing infrastructure and has partnered with the IDRC hearing venue in London on a dedicated suite, which integrates with a venue rather than a system. The platform page was not opened and may name connectors.

Residency is offered explicitly, and processing location is not addressed. The security page says the vendor hosts each client's solution and data in the location that best fits its needs, which is a customer choice, supported by separate EMEA, North America and Asia Pacific operations and a list of data protection regimes the services meet: GDPR, UK GDPR, Canadian PIPEDA, the California Consumer Privacy Act, the Singapore Personal Data Protection Act and the Australian Privacy Act.

Data is described as self contained through firewalls, encrypted volumes, logical identifiers and ownership tags, which is logical rather than physical separation. No list of regions or data centers is published, no cloud provider is named, and nothing separates where data is stored from where processing or model inference happens, which matters to a customer choosing a jurisdiction to keep material inside it.

Current, named certification is published with nothing to verify it. The security page, last modified 19 Aug 2026, names ISO/IEC 27001:2022 and Cyber Essentials Plus, with regular CREST accredited external penetration testing, STRIDE threat modeling with risk owners, and automated and manual code review. It cites the 2022 revision of ISO 27001 rather than the superseded 2013 version. No certification body, certificate number, scope, validity period or issue date is given, there is no route to obtain the certificate or a report, no trust portal, and no SOC 2 of any type, which US buyers look for first.

The arbitral institutions page states ISO 27001:2013 while the security page states 2022, so the site gives two versions of the same certification, one out of date.

The vendor names its AI and never says what runs it. Opus 2 Ingenuity is described as an agentic framework working on matter data, covering summaries, entity extraction, questions, prompt building, drafting and real time speech recognition. Searched the AI, home, security and arbitral institutions pages: no model or provider is named, no first party versus third party distinction is drawn, no version appears, inference location is not stated beyond the general offer to host in a chosen geography, and there is no commitment to notify customers of model or configuration changes.

The one related commitment is that client data never leaves the client's instance, which says where data does not go without saying what processes it. For AI that reads live testimony in a courtroom, the model is what a tribunal would most likely ask about.

No pricing is published at any level, including the unit of charge. The full navigation and footer across all three regional sites, read, have no pricing or plans page. No rate, unit, tier, volume band, term or charge for implementation, hearing services or transcription appears, which matters because the offering mixes software with services such as court reporters, transcripts and hearing room setup, each usually priced differently.

Every call to action is to watch a demo or make contact. The only cost related claim is that connected data and workflows make AI use more efficient, so customers spend less.

Coverage is documented by role and by solution, and no boundary is stated. Role pages cover partners and lead counsel, legal innovation teams, associates, and paralegals and litigation support, each with its own benefits. Solutions cover Cases, Hearings, Transcripts, Arbitral Institutions, and Chambers sold as LEX, identifying law firms, barristers' chambers and arbitral institutions as buyers. Practice focus is commercial disputes, international arbitration, public inquiries and inquests, with named inquiry work and arbitral institution deployments, and the chambers product reaches criminal clerking.

Reach spans three regional operations and six data protection regimes. The vendor says it serves the litigation and arbitration lifecycle in any jurisdiction worldwide, and nothing names a practice area, matter type, jurisdiction or scale it does not serve.