Corlytics vs Norm Ai: how they compare in 2026

Machine learning does three jobs on every incoming document: it decides what the document is about, condenses it, and connects it to related material. Regulatory notices, rules and enforcement actions are classified against Corlytics' universal regulatory taxonomy, long texts are summarized, and similar items are matched across datasets to flag duplication and feed gap analysis. Obligations Management goes further. Semantic parsing strips introductory and suggestive text from a rule to leave a set of action items. Specialized models map those obligations to a firm's policies, controls and products and consolidate the overlaps, and large language models propose wording that would close a gap in a policy or control. Policy Management drafts new policy content with AI, and Regulatory Risk Analytics runs Corlytics' own risk models over enforcement data. The machine learning framework that tags handbook provisions in real time was first built with the FCA in 2016 and 2017. Without the models, the version tracking, change packages, approval workflows, controls mapping and monitoring plans would still run as a regulatory workflow system.

The agents are the product. Legal Engineers encode regulations and firm policies into decision trees written in Norm's own domain specific language. An agent works through the tree to decide whether content complies and, if not, why. Content review, questionnaire drafting and the compliance agent in Microsoft 365 Copilot all run on that agent layer. Norm Law, the firm Norm powers, delivers its legal work through the same platform. Without the models, what remains is a library of encoded rules with nothing to apply them.

Every alert, summary and obligation stays linked to the regulatory text it came from. The Regulation Library keeps laws, rules, handbooks and guidance in their original PDF, XML or HTML form, so second and third line teams read the documents they already know. It breaks each one down to obligation level. A version manager shows the current text of a rule beside the previous version, with changes lined in red and green, and keeps the full history. Obligations mapped to policies and controls point back to that source. Quality control is described as a process rather than a number. Subject matter experts at Corlytics validate the enriched content, and in Obligations Management the client's own experts review extracted and mapped obligations before gap reports go out. In the FCA handbook project, machine tagging ran with a review, approve or reject step and a full audit trail. No accuracy rate or test result is given for classification, summaries, extraction or mapping.

Each determination comes with an explanation tied to the rule that produced it. A reviewer can see why content was flagged and what would make it compliant. Because the reasoning path is the encoded regulation itself, a finding traces to a provision rather than to a retrieved passage. Questionnaire answers are cited to firm approved sources, and reviewers can jump straight to the supporting excerpt. Each answer is labeled as coming from the Answer Bank, generated by AI or needing action. Prudential's leaders describe accuracy close to 100% on the most important issues after calibration. That account gives no test set, sample or method, and Norm publishes no accuracy figure or error rate of its own.

The platform automates the reading and routing of regulation and leaves decisions to people. Content is collected, classified and sent to the right business lines and teams without manual triage. Extraction and mapping run automatically, then the client's subject matter experts validate the results before gap reports and attestations are produced. Gap analysis returns suggested wording for a policy or control rather than changing it, and users are prompted to revisit mappings when a rule changes. Change management turns a regulatory change into a change package with owners, tasks and alerts at each stage, and records offline gap analysis alongside it. Policies pass through configurable multistage approvals. For the FCA's handbook, machine tagging ran inside a four eyes workflow in which users could review, approve or reject each change. No confidence level is described at which the models hold back output, and nothing covers what happens once a wrong mapping has reached a policy or control.

Agents do the first review and a person decides. At New York Life the agents give reviewers recommendations tied to several regulatory frameworks, and compliance professionals make the final decisions. Prudential's marketers screen content in Norm before it reaches compliance, and reviewers finish the check on the same platform. Questionnaire drafts move through the firm's own sign off by business unit, legal, compliance and management. Questions without enough support are flagged for a reviewer with the reason given. A change to an encoded rule needs review by a second certified Legal Engineer. Nothing published sets a point at which a content review agent stops, or says what follows when an approved item later proves wrong.

The longest running named deployment is the FCA. A pilot began in September 2016 and the Intelligent Handbook went live on 10 May 2017, with Corlytics' taxonomy and machine learning tagging covering more than 18,000 provisions. Other FCA functions use it too, including to speed up firm onboarding. Arbuthnot Latham, a London private and merchant bank, is named in a case study about centralizing group policies, with standard templates, single sign on and staff attestation among its requirements. The International Compliance Association launched an alert tool on Corlytics feeds in June 2025 for more than 30,000 members, and GRC Report began carrying its feeds in September 2026. Six more case studies are anonymous. Corlytics puts the efficiency gain at 25% for the compliance monitoring and risk insight teams of a European tier 1 bank. A crypto exchange went live in six weeks, and a United States investment manager with more than one trillion in assets replaced older third party tools with it. Client logos also include FINRA, ING, abrdn, Block, Natixis, Patria, SCOR, Swiss Re and Vontobel. The headline claims, 40% of the world's 30 largest banks as clients and policy approvals 50% faster, come without a named source.

Two named insurers describe their use. New York Life's compliance team uses Norm for sales and marketing review across several business units. Its chief compliance officer, Sandi Tillotson, is quoted in Norm's October 2025 announcement, and NYL Ventures is also an investor. Prudential's Director of Generative AI and its head of marketing strategy described their rollout in January 2026. Marketers screen content in Norm before compliance sees it, and the rollout began with the content most affected by generative AI. Prudential says its content volume rose five to ten times, and its remark on accuracy carries no method. Neither account gives a before and after figure for review time. Blackstone is named as a client of Norm Law.

Each client's policies, obligation registers, impact assessments and control mappings sit in the platform, and the terms that govern that content are not public. No customer agreement, data processing agreement or confidentiality schedule is published. Nothing public says how client content is used, whether it trains the models, or how one client's data is kept apart from another's. The June 2026 privacy notice is the only data document. It covers personal data about people outside the customer relationship, such as website visitors, prospects and suppliers, though its reader list also names software users. It describes access controls, user authentication, encryption, staff confidentiality and regular security testing. ISO/IEC 42001 certifies how Corlytics governs its own AI, and ISO/IEC 27001 covers its information security management. Neither is a statement to clients about how their content may be used.

Norm's commitments here are security commitments. Customer data is encrypted in transit with TLS 1.2 or later and at rest with 256 bit encryption. Access follows least privilege with quarterly reviews, and an annual SOC 2 Type II audit covers the controls. No published term says whether customer content may train or improve models. Nor is there a statement on how one customer's material is kept apart from another's, or how privileged and work product material is handled. The privacy policy lets Norm use data to build and improve its Site and Services. The model providers that process inputs are named, though their retention terms are not. The platform reviews draft marketing, investor material and internal communications before release, so it holds content while a compliance problem is still private.

The product sells interpretation as well as information. The platform extracts obligations, filters them to a firm's activities and jurisdictions, suggests how a policy or control should change, and drafts policy text. In one UK bank's case, a third party consultancy used rules supplied by Corlytics to complete a rules mapping exercise the bank had agreed with its conduct regulator. The website legal notice says site content is general guidance that does not replace regulatory, compliance, risk analytics or other professional advice, and supplies it as is. It governs the website rather than the platform. No public statement says how platform output relates to legal or compliance advice, or who is responsible for checking it before a firm relies on it. The client expert validation step in Obligations Management is the clearest description of who signs off.

Norm separates the software from legal advice by structure, and says so in writing. Norm Law's notices state that Norm Ai is a technology company and not a law firm, and that it gives no legal advice. All legal services come from Norm Law's licensed attorneys, under a services arrangement with Norm. An attorney client relationship with Norm Law begins only with a signed engagement letter. On the compliance side, agents make recommendations and compliance professionals decide. Norm Law is chaired by Mike Schmidtberger, formerly chair of Sidley Austin's executive committee, and Norm's experts include Troy Paredes, a former SEC Commissioner. Nothing published addresses a lawyer's competence or supervision duties when using the agents, or names jurisdiction limits.

An AI management system certified to ISO/IEC 42001:2023 covers the whole business. A-LIGN issued the certification in May 2025 after a two stage audit that found zero nonconformities, and Waystone Compliance Solutions advised on the program. The scope covers every product line. The program is framed around bias mitigation, risk and resilience controls across the AI lifecycle, and mapped to the EU AI Act and the UK National AI Strategy. Later releases add the NIST framework and OECD principles. Oisin Boydell, Chief Data Officer, speaks for the AI program, and founder John Byrne describes an AI by design strategy that reaches six competencies of the business. Product pages describe a science based method, ensemble models and expert validation at each stage. Test results are not published, so nothing shows how output quality varies across languages, jurisdictions or document types.

Governance is built into how agents are made. Legal Engineers encode each regulation into a readable decision tree. A change to an encoded rule needs review by a second certified Legal Engineer. Every determination carries an explanation a lawyer can audit without reading code. Norm says its evaluation work ties agent performance to the quality, time and cost of the result. It also builds supervisory agents that monitor other AI systems against laws and policies. No evaluation results, named owner for model governance or AI management certification are published. Nothing addresses uneven output across content types or populations.

Clients' regulatory content, obligation registers and policies live in Corlytics' cloud platform, and no published document sets retention, deletion or subprocessor terms for them. The privacy notice covers personal data only. Corlytics keeps that data for the life of an active contract, or as legal and business needs require, then deletes or anonymizes it, with the detail in an unpublished Data Retention Policy. Breaches are notified without undue delay and reported to the supervisory authority under an unpublished Data Breach Management Policy. Luca Dalla Giacoma is the named Data Protection Officer. The same notice lists access controls, authentication, encryption, IP anonymization, network security, business continuity and disaster recovery, and regular testing. It names recipients only by category: staff, technical service providers acting as processors or subprocessors, authorities such as HMRC, professional advisers and buyers of the business. The ISO/IEC 27001 certificate covers the information security management behind the service.

Norm's trust center and security fact sheet cover most of the data ground. Data is encrypted in transit and at rest. The platform runs on cloud providers such as Amazon Web Services, with production isolated behind network segmentation and firewall rules. Access follows least privilege, with quarterly reviews, multifactor sign in for staff and SAML single sign on for customers. A written incident response plan commits Norm to contain an incident and notify affected parties as appropriate. Annual penetration tests and an annual SOC 2 Type II audit check the controls. The subprocessor list names Anthropic, OpenAI, Microsoft Azure and Baseten for model inference, and Reducto AI for document conversion. Retention procedures exist and customer data is deleted when a customer leaves, but no retention period is published.

The only public legal document is a website notice. It supplies site content as is, disclaims accuracy, timeliness and completeness, and excludes liability for direct, indirect, consequential and other damages arising from the site. The client application sits behind a sign in, and no customer agreement, order form or service terms is published, so the platform's warranties, indemnities and liability limits are not public. Clients use the output in front of regulators. Firms rely on Corlytics mappings to evidence compliance and attest to supervisors, and one UK bank used Corlytics rules in a rules mapping exercise agreed with its conduct regulator. No public term says who carries the loss if a mapping, summary or suggested policy change turns out to be wrong.

No customer agreement is published. The site terms of January 2024 govern the website and its content. They cap Norm's liability at $100 and disclaim the accuracy of site material. Use of the services is left to separate user terms that are not public. Nothing published says who carries the loss when a compliance determination is wrong, for example when cleared marketing later draws a regulatory finding. Norm Law carries its own professional responsibility as a law firm, which does not extend to the software.

The platform is built to sit beside a firm's GRC stack and feed it. The Regulation Library is described as fully API ready, with an Open API for major GRC platforms. Regulatory Monitoring, Obligations Management and Compliance Monitoring each send data or outcomes into a firm's GRC environment, and an investment manager client uses the data to feed its internal compliance newsletter. The FCA handbook project connected to the regulator's existing content repository, and Taxonomy Manager for Regulators links digital handbooks to online portals and obligation registers. The partners page shows logos for IBM, ServiceNow, Solidatus and Chainalysis among others. Inside the platform, controls mapping links obligations to a firm's policies, standards, procedures, controls and processes, and taxonomy routing sends content to the right business lines and teams. No connector is named, and there is no public API reference or integration guide that shows what syncs and in which direction.

Norm works inside Microsoft 365. In Word and PowerPoint its agents review a draft in place. They flag missing disclosures, unsupported claims and policy conflicts, each tied to the text or slide concerned, so teams fix them without leaving the file. A compliance agent in Microsoft 365 Copilot reviews content, checks key facts against approved sources, answers policy questions and keeps an audit trail. Norm is listed on the Microsoft Marketplace. Questionnaires export back to their original format, or to PDF or Excel, with citations and version history. No connector documentation describes setup, what syncs or in which direction. No archive, CRM or document management connection is named.

Delivery runs from Corlytics' own cloud, the environment named in its ISO/IEC 27001 scope. Implementations run remotely across time zones, led by a dedicated implementation team, and Corlytics says go lives can take a matter of weeks. The security page lists geo specific hosting among its principles. The hosting provider, regions and tenancy model are not named, and no choice of location is offered. The only geography in the privacy notice concerns personal data transfers outside the UK and EEA, which rely on adequacy decisions, approved contracts or other safeguards. Group companies in Ireland, England, Portugal and the United States describe where Corlytics operates rather than where client data sits.

Norm is delivered as cloud software. Its security fact sheet says the cloud environments run on providers such as Amazon Web Services. Model inference runs through Anthropic, OpenAI, Microsoft Azure and Baseten. No hosting region, residency choice or tenancy model is published. Nothing separates where data is stored from where models process it. The buyers are regulated financial institutions, for whom data location is often a supervisory question.

Three certifications are named. ISO/IEC 27001 covers all activities of Corlytics Limited in delivering data and software for legislation, regulation and document management services from its cloud solutions. A SOC 2 Type 1 report from the independent CPA firm Thoropass tests control design against all five Trust Services Criteria at a point in time. ISO/IEC 42001:2023 came from A-LIGN in May 2025. An Information Security Manager runs the security program, whose principles include secure development, testing and quality assurance, continuous integration testing, infrastructure controls, geo specific hosting and incident response. One reinsurer cleared Corlytics through its own digital governance process as an approved third party supplier before rollout. No report or certificate is posted, and the security page invites readers to get in touch for more on its procedures.

Norm completes an annual SOC 2 Type II audit against the AICPA Trust Services Criteria. Its trust center, run on Vanta, lists the SOC 2 Type II report, a SOC 2 bridge letter, a penetration test summary for the first half of 2026, a business continuity exercise letter and CAIQ responses. Each sits behind a request for access, and Norm says the reports are available to customers. The security fact sheet and the chief security officer's profile open without a request. No audit period, auditor or scope is shown in public, and no other certification is held. Norm says further certifications are on its roadmap.

The AI is presented as the company's own work. Ensemble models with expert oversight handle classification and summaries, and specialized machine learning models handle obligation mapping. Gap remediation suggestions come from what Corlytics calls expert LLMs, and handbook tagging runs on a machine learning framework first built with the FCA. No model, model family or provider is named for any of them, and whether a third party foundation model is called, or where inference runs, is not stated. The ISO/IEC 42001 management system covers every product line, with no list of the models it governs and no commitment to tell clients when they change.

Norm's subprocessor list names the companies behind its models. Anthropic, OpenAI and Microsoft Azure supply language model infrastructure. Baseten hosts and runs models and processes the inputs sent to them. Reducto AI turns uploaded PDFs, spreadsheets and slides into text the models can read. The regulatory logic is Norm's own, encoded by Legal Engineers in its domain specific language, and a general purpose model executes against it. No model versions are named, and nothing says which provider serves which feature or where inference runs. No commitment to notify customers of a change is published.

Buying Corlytics starts with a demo and a sales conversation, and no prices, plan names or units of charge are published. The platform is modular. In its case studies, clients start with Regulatory Monitoring and add Risk Analytics, Compliance Management or Clausematch policy management, and one payments provider added two modules within a month of going live. Implementation is run by a dedicated team, with an implementation manager for each client, in sprints. A crypto exchange went live in six weeks, while one reinsurer's rollout ran through requirements and design, a proof of concept and approval as a third party supplier. Modules and configuration can be added after go live. No figure is published for licenses, implementation or the work of building a firm's taxonomy and obligation register.

No price, plan or unit of charge is published. Norm's sites, Norm Compliance included, end in a contact form. Its Microsoft Marketplace announcement gives no price either. No free trial or self serve entry exists. The buyers are large regulated institutions, and every purchase runs through Norm's own team.

Buyers are regulated financial firms and the regulators that oversee them, with a second line of business in health and life sciences. Sector pages cover global and private banks, hedge fund and investment managers, global insurers, payments and big tech, and regulators. Case studies add a reinsurer, a crypto exchange, a UK retail and commercial bank, a London merchant bank and a United States investment manager. Regulators buy a Taxonomy Manager edition that turns handbooks into machine readable, searchable databases with visual maps by theme, product and business line. Health and life sciences clients get regulatory sensing, guidance assessment, digital call for comments assessments and a repository that supports submissions. Financial services modules cover horizon scanning, the regulation library, obligations, change management, controls mapping, policy management, compliance monitoring, regulatory risk analytics and nonfinancial risk across operational, conduct, third party, cyber, AI and governance risk. Product pages speak to compliance, risk and audit teams and their chief officers, and case studies show legal teams working alongside them. No limit by firm size, jurisdiction or language is stated.

Norm serves large regulated institutions, chiefly financial services and insurance organizations. It says its clients represent more than $35 trillion in assets under management. Chief compliance officers and their teams are the main users. Encoded sources include SEC, FINRA and CFPB rules alongside each firm's own policies. Workflows cover regulated content review in marketing, sales, investor relations and internal communications, along with due diligence questionnaires and RFPs. Supervisory agents monitor other AI systems. Legal advice sits with Norm Law, whose practice areas run from private funds and private equity to tax and to FINRA and New York financial regulation. No coverage outside the United States, segmentation by size or list of what the platform is not built for is published.