AI oversight at board level and an approval process, but no test evidence for its own AI
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Marsh & McLennan assigns AI oversight to the Audit Committee and describes a three-step AI Risk Framework, but a policy, metrics and external audit are missing.
What is documented
- The 2026 proxy statement assigns oversight of AI and cyber risks to the Audit Committee; the 2025 version did not yet mention AI1 2
- The AI Risk Framework requires screening, expert review and approval by Business AI Committees for all new AI projects3 5
- Only approved AI tools are permitted; confidential data must not be entered into unapproved systems3
- Recruitment is treated as a high-risk area in the framework; human oversight is required where there is a risk of bias or error3
- More than 46,000 employees have completed the AI Academy since August 20248
- An ethics and compliance hotline allows anonymous reports and prohibits retaliation3 9
The case against
Self-description without figures
- The AI policy and the AI Risk Framework are not published in full; figures on reviewed or rejected projects are missing
- The Business Responsibility Reports are not externally assured; no certification of AI management was found3 5
- Formal risk tiers and a separate process for purchased AI are not described3
Gaps in responsibility and complaints
- No named person responsible for AI governance is disclosed
- An AI-specific complaints channel and documented remedy are missing
Sources for Responsibility (9)
- Marsh & McLennan Companies, Inc. / U.S. SEC: Definitive Proxy Statement 2026 (DEF 14A) · 8 April 2026 · primary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: Definitive Proxy Statement 2025 (DEF 14A) · 3 April 2025 · primary source
- Marsh: 2025 Business Responsibility Report «From Insight to Impact» · 31 December 2025 · primary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: 10-K 2025, Cybersecurity Risk Management and Strategy Disclosure (XBRL view R29) · 9 February 2026 · primary source
- Marsh McLennan: Perspective to Thrive: 2024 Business Responsibility Report · 31 December 2024 · primary source
- Marsh McLennan Netherlands: Perspective to Thrive: 2024 Business Responsibility Report, addendum Marsh McLennan Netherlands · 2025 · secondary source
- Marsh McLennan: Marsh McLennan develops new generative AI tool (press release on LenAI) · 21 November 2023 · secondary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-31 · 9 February 2026 · primary source
- Marsh & McLennan Companies, Inc.: The Greater Good – Code of Conduct (© 2021–2024) · 2024 · secondary source
2 · Safety
The group documents a security programme aligned with the NIST framework, with external reviews and SOC 2 reports, but discloses no testing, validation or bias checks of its own AI systems.
What is documented
- Item 1C documents a NIST-based programme with a CISO, Audit Committee oversight, external reviews, penetration tests and supplier assessment4
- SOC 2 Type II audits cover all data centres and the cloud infrastructure; testing follows MITRE ATT&CK1 5
- LenAI uses a privately hosted version of OpenAI models via Microsoft with minimal data retention6 7
- Mercer commits to human review of all AI-supported outputs in its aptitude assessments2
- Formal incident procedures provide for escalation to executive management and the board4
The case against
Missing evidence of AI testing
- No test reports, validation studies or bias audits on LenAI, Claims IQ or Mercer's assessments were found; the “conformity testing” mentioned is not explained2
- Item 1C does not mention AI; AI-specific controls such as prompt and model protection are not described4
- ISO/IEC 27001 is documented only for Mettl and only as a company statement8
Data incidents
Sources for Safety (12)
- Marsh: 2025 Business Responsibility Report «From Insight to Impact» · 31 December 2025 · primary source
- Mercer (Marsh): Responsible use of AI in talent assessments · 3 November 2026 · secondary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-31 · 9 February 2026 · primary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: 10-K 2025, Cybersecurity Risk Management and Strategy Disclosure (XBRL view R29) · 9 February 2026 · primary source
- Marsh McLennan: Perspective to Thrive: 2024 Business Responsibility Report · 31 December 2024 · primary source
- Marsh McLennan: Marsh McLennan develops new generative AI tool (press release on LenAI) · 21 November 2023 · secondary source
- Fortune (John Kell): Marsh McLennan's CIO started with a summer job in 1995 ... · 27 November 2024 · secondary source
- Mercer | Mettl (Marsh): Information Security Features · 3 November 2026 · secondary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: Annual Report to Shareholders 2025 (Form ARS) · 8 April 2026 · secondary source
- Mealey's (LexisNexis): 1 Plaintiff, 1 Claim Remain In Marsh & McLennan Data Breach Class Action · 10 June 2025 · secondary source
- Marsh & McLennan Companies, Inc.: The Greater Good – Code of Conduct (© 2021–2024) · 2024 · secondary source
- National Consumer Law Center (Bibliothek): Bohnak v. Marsh & McLennan Cos. (2d Cir., decision of 2023-08-24) · October 2025 · secondary source
3 · Transparency
The group describes its AI use and the model behind LenAI only in general terms and publishes neither an inventory of AI systems nor system cards or information on data sources.
What is documented
- Mercer commits to informing clients and candidates about AI involvement and labelling it in reports1
- Mercer states that AI results are diagnostic, not prescriptive, and make no final decisions1
- LenAI mainly uses OpenAI models via Microsoft Azure; there is no large-scale training of its own models6 8
- The risk factors in the 10-K disclose AI risks such as flawed model design3
The case against
Missing system information
- No inventory of AI systems in use, no system cards and no transparency report were found
- For Aida, Workforce Insights and Sentrisk there is no information on limitations, explainability and data sources
- Mercer's commitments to inform candidates are not externally verified1
Sources for Transparency (8)
- Mercer (Marsh): Responsible use of AI in talent assessments · 3 November 2026 · secondary source
- Marsh McLennan: Marsh McLennan develops new generative AI tool (press release on LenAI) · 21 November 2023 · secondary source
- Marsh & McLennan Companies, Inc. / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-31 · 9 February 2026 · primary source
- Mercer / Business Wire: Mercer Embraces AI to Deliver Better Outcomes for Clients; Launches AI Talent Capabilities, Workforce Insights and Aida · 2 October 2025 · secondary source
- Marsh McLennan: Marsh McLennan launches AI-powered solution to transform supply chain risk management (Sentrisk) · 2 May 2024 · secondary source
- Fortune (John Kell): Marsh McLennan's CIO started with a summer job in 1995 ... · 27 November 2024 · secondary source
- Marsh McLennan Netherlands: Perspective to Thrive: 2024 Business Responsibility Report, addendum Marsh McLennan Netherlands · 2025 · secondary source
- Marsh: 2025 Business Responsibility Report «From Insight to Impact» · 31 December 2025 · primary source
4 · Why the harm rule does not apply
The harm rule does not apply: in the review period there is neither a judgment nor an authority decision nor a fine concerning its AI use. The Bohnak data privacy lawsuit is minor and has no AI reference.
5 · Controversies (1)
6 · What remains open
- Figures on the AI Risk Framework, such as reviewed and rejected AI projects and high-risk cases
- Independent audit of AI management or of the Business Responsibility Reports
- Test, validation and bias results of Mercer's aptitude assessments
- Current status of the Bohnak class action
- Items 1C and 3 of the 10-K in the main HTML text, which were truncated on retrieval
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 24 fetches.
Primary documents read (7)
- Form 10-K fiscal year 2025 (Item 1, Item 1A; two parts) and XBRL exhibits R29 (Item 1C) and R25 (Legal Note)
- Annual Report to Shareholders 2025 (Form ARS, two parts, up to Item 1A)
- Definitive Proxy Statements 2025 and 2026 (first 100,000 characters each)
- Form 8-K relating to the AGMs 2024 and 2025
- Business Responsibility Reports 2024 and 2025 and Addendum Netherlands 2024
- Code of Conduct «The Greater Good» (© 2021–2024)
- Mercer «Responsible use of AI in talent assessments» and Mettl «Information Security Features»
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0