Clear AI responsibilities, but errors in its legal AI and lawsuits over personal data
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
The annual report documents a chain of responsibility from the Responsible AI Working Group up to the CEO, but there is no group-wide approval procedure, no AI metrics and no external audit.
What is documented
- The CEO is accountable to the board for corporate responsibility including the Responsible AI Principles2
- The Chief Technology Officers are responsible for the principles, and a Responsible AI Working Group provides oversight3
- The board has reserved risk appetite, risk oversight and corporate responsibility activities to itself1
- The risk report lists fabricated content from generative AI as an emerging risk with countermeasures such as testing and source-linked citations5
- Periodic algorithmic impact assessments are described for Scopus AI6
The case against
Gaps in governance and review
Sources for Responsibility (12)
- RELX PLC: Annual Report 2025 – Governance (section) · 19 February 2026 · primary source
- RELX PLC: Annual Report 2025 – Corporate Responsibility (section) · 19 February 2026 · primary source
- RELX PLC: Annual Report 2025 – Sustainability Statement and other Corporate Responsibility Disclosures · 19 February 2026 · primary source
- RELX: RELX Responsible AI Principles · 1 June 2022 · secondary source
- RELX PLC: Annual Report 2025 – Financial Review, Principal and Emerging Risks (section) · 19 February 2026 · primary source
- Elsevier: Scopus AI – Fact Sheet: Using Responsible AI Principles · 1 March 2024 · primary source
- Elsevier: Responsible AI Principles (Elsevier) · 3 November 2026 · secondary source
- LexisNexis Risk Solutions: AI Principles Booklet 2025 · 3 November 2026 · secondary source
- LawSites (LawNext): The Legal Tech Giants Powering ICE, Part 2 — The Pushback · 29 April 2026 · secondary source
- RELX: Corporate Responsibility 2025 (web edition) · 31 December 2025 · secondary source
- Nieman Lab: Law360 mandates reporters use AI «bias» detection on all stories · 1 July 2025 · secondary source
- Gewerkschaftlich organisierte Mitarbeitende von Law360 und MLex (veröffentlicht via Poynter): Open letter to the RELX CEO, CFO and board on the ICE contract · 3 November 2026 · secondary source
2 · Safety
System protection is documented in the 20-F and by third-party audits, but a Stanford study found hallucinations in Lexis+ AI, and two data breaches and proceedings over personal data make “solid” a borderline judgement.
What is documented
- The 20-F describes an Information Security Council with semi-annual risk assessments, board briefings and internal audit reporting to the Audit Committee9
- More than 60% of product revenue of the three largest divisions is covered by third-party audits, UK Risk products under ISO 270012
- Scopus AI has a quality and bias assessment framework for each release1
- ClinicalKey AI has an assessment framework developed with clinical experts2
- According to the Trust Center, customer data is not used for AI training11
The case against
Independent findings on legal AI
Incidents and proceedings on personal data
- 2024 data breach at LNRS affecting 364,333 people, followed by a class action; 2026 intrusion at Legal & Professional12 13 14
- Settlement of USD 13.5 million over false death records in identity checks, without any finding8
- According to the lawsuit, GM driving data was obtained without consent and turned into driving scores; proceedings pending17
Sources for Safety (19)
- Elsevier: Scopus AI – Fact Sheet: Using Responsible AI Principles · 1 March 2024 · primary source
- RELX PLC: Annual Report 2025 – Corporate Responsibility (section) · 19 February 2026 · primary source
- LexisNexis: Humans in the Loop: The People Powering Trusted Legal AI · 30 January 2026 · secondary source
- Stanford HAI: AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries · 23 May 2024 · primary source
- Magesh, Surani, Dahl, Suzgun, Manning, Ho (Stanford RegLab/HAI), arXiv: Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arXiv 2405.20362) · 30 May 2024 · primary source
- Legal IT Insider: Canadian law professor gives Lexis+ AI «a failing grade» – LexisNexis responds · 18 November 2024 · secondary source
- LawSites (LawNext): Vals AI's latest benchmark finds legal and general AI now outperform lawyers in legal research accuracy · 23 October 2025 · secondary source
- ClassAction.org: $13.5m LexisNexis Settlement Ends Class Action Lawsuit Over Allegedly Wrongful Death Reports · 2 December 2025 · secondary source
- RELX PLC / U.S. SEC: 20-F 2025, Cybersecurity Risk Management and Strategy Disclosure (XBRL view R37) · 19 February 2026 · primary source
- RELX PLC: Annual Report 2025 – Governance (section) · 19 February 2026 · primary source
- LexisNexis: The LexisNexis Trust Center: Building Confidence Through Transparency · 14 November 2025 · secondary source
- ClassAction.org: LexisNexis Risk Solutions Data Breach · 28 May 2025 · secondary source
- Bloomberg Law: LexisNexis Sued Over April Data Breach Affecting 364,000 People · 29 May 2025 · secondary source
- The Register: LexisNexis Legal & Professional confirms data breach · 4 March 2026 · secondary source
- BleepingComputer: LexisNexis confirms data breach as hackers leak stolen files · 3 March 2026 · secondary source
- LexisNexis Risk Solutions: AI Principles Booklet 2025 · 3 November 2026 · secondary source
- U.S. District Court for the Northern District of Georgia: In re Consumer Vehicle Driving Data Tracking Collection, MDL No. 3115, No. 1:24-md-03115-TWT, Opinion and Order on Motions to Dismiss (Doc. 203) · 22 April 2026 · primary source
- TechCrunch: GM banned from sharing driving and location data with insurance companies · 17 January 2025 · secondary source
- Bloomberg Law: Immigration Groups Drop LexisNexis Suit Over ICE Info Sharing · 16 May 2024 · secondary source
3 · Transparency
RELX publishes its AI principles and discloses the model and methods for Scopus AI, but for its other products there are no system or model cards, transparency reports or error rates.
What is documented
- RELX publishes its Responsible AI Principles, with separate versions from Elsevier and LexisNexis Risk Solutions1 9 12
- The annual report describes the divisions' AI products and AI risks3 4
- For Scopus AI, the model, hosting via Azure, retrieval method and answer tiers by source confidence are disclosed7
The case against
Missing product information
- No system or model cards and no error rates were found for Lexis+ AI, ClinicalKey AI and the Risk models
- That the Law360 tool is based on GPT-4 only became known through media reports5
Sources for Transparency (12)
- RELX: RELX Responsible AI Principles · 1 June 2022 · secondary source
- RELX: Responsible AI Principles (website) · 3 November 2026 · secondary source
- RELX PLC: Annual Report 2025 – Financial Review, Principal and Emerging Risks (section) · 19 February 2026 · primary source
- RELX PLC: Annual Report 2025 (ESEF HTML, UK National Storage Mechanism) · 19 February 2026 · secondary source
- Nieman Lab: Law360 mandates reporters use AI «bias» detection on all stories · 1 July 2025 · secondary source
- Retraction Watch: Elsevier denies AI use in response to evolution journal board resignations · 6 January 2025 · secondary source
- Elsevier: Scopus AI – Fact Sheet: Using Responsible AI Principles · 1 March 2024 · primary source
- Legal IT Insider: Canadian law professor gives Lexis+ AI «a failing grade» – LexisNexis responds · 18 November 2024 · secondary source
- Elsevier: Responsible AI Principles (Elsevier) · 3 November 2026 · secondary source
- Magesh, Surani, Dahl, Suzgun, Manning, Ho (Stanford RegLab/HAI), arXiv: Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arXiv 2405.20362) · 30 May 2024 · primary source
- LawSites (LawNext): Vals AI's latest benchmark finds legal and general AI now outperform lawyers in legal research accuracy · 23 October 2025 · secondary source
- LexisNexis Risk Solutions: AI Principles Booklet 2025 · 3 November 2026 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: in the period there is no judgment, no authority decision and no penalty concerning harm from the group's AI. The settlement over false death records is not a judgment, and the driving data proceedings have produced no finding.
5 · Controversies (9)
- mediumAccording to the lawsuit, LexisNexis Risk Solutions obtained GM driving data without consent and sold driving scores derived from it to insurers; dismissal granted only in part. · 22 April 2026 · open
- mediumLexisNexis Risk Solutions agreed a USD 13.5 million settlement over false death records for around 59,000 living people. · 4 November 2025 · Vergleich
- mediumAfter a December 2024 data breach affecting 364,333 people, a class action was filed against LexisNexis Risk Solutions. · 29 May 2025 · open
- minorLexisNexis Risk Solutions provides ICE and CBP with database and analytics access; in 2026 more than 200 of the group's journalists called for an exit. · 29 April 2026 · open
- minorLexisNexis Legal & Professional confirmed an intrusion into a limited number of servers holding mostly legacy data; a larger scope is claimed only by the attackers. · 3 March 2026 · unklar
- minorLaw360 required its newsroom to use a GPT-4-based bias tool; more than 90% of union members called for it to be voluntary. · 1 July 2025 · unklar
Further controversies (3)
- minorThe editorial board of the Journal of Human Evolution accused Elsevier of undisclosed AI use; Elsevier denies this. · 26 December 2024 · unklar
- minorA Stanford study found around 17% hallucinations in Lexis+ AI, although the product was marketed with hallucination-free citations. · 23 May 2024 · unklar
- minorA lawsuit against LexisNexis Risk Solutions over data sales to ICE and others was withdrawn after a motion to dismiss was granted. · 15 May 2024 · abgewiesen
6 · What remains open
- Accuracy and hallucination rates of the current versions of Lexis+ AI and Protégé
- Group-wide approval procedure and risk classification for AI systems
- Human rights and misuse review for government customers of LexisNexis Risk Solutions
- Status of the class action over the 2024 data breach and of the driving data proceedings
- Scope of the external assurance of security and data protection metrics
- Revised Responsible AI Principles 2026 (in progress according to the annual report)
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 37 fetches.
Primary documents read (6)
- Form 20-F 2025 (two parts) and XBRL presentation R37 (Item 16K), filing index and FilingSummary
- Annual Report 2025: sections on Governance (two parts), Financial Review including Principal Risks, Corporate Responsibility, Sustainability Statement; ESEF HTML (two parts)
- RELX Responsible AI Principles (PDF) and website; Elsevier Responsible AI Principles; LNRS AI Principles Booklet 2025
- Opinion and Order on Motions to Dismiss, MDL No. 3115 (N.D. Ga., 2026-04-22)
- Stanford study arXiv 2405.20362 and HAI summary
- RNS Result of AGM 2026
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0