Tested diagnostic platforms, but no documented AI oversight at board level
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Executive management has adopted AI ethics principles, but the board committees mention neither AI nor data nor cybersecurity.
What is documented
- Executive management adopted the AI ethics principles on 20 December 2023 and updated them in 20251
- The principles require accountability across the entire life cycle, including third-party systems, and human oversight1
- Data ethics principles and an AI clause in the code of conduct supplement the AI principles11 6
- The Chief Digital and Technology Officer has been a member of the enlarged Corporate Executive Committee since 20252 3
- Speak-up channels via an external provider allow anonymous reports and escalate up to board committees6
The case against
Oversight not documented
Sources for Responsibility (11)
- F. Hoffmann-La Roche Ltd: Roche Artificial Intelligence (AI) Ethics Principles · 1 April 2025 · secondary source
- Roche Holding AG: Annual Report 2025 (English) · 29 January 2026 · secondary source
- Roche Holding AG: Wafaa Mamilli (leadership profile) · 3 November 2026 · secondary source
- CIO.com (Martha Heller): How Roche drives AI outcomes in a changing world (interview with Wafaa Mamilli) · 16 September 2026 · secondary source
- Roche Holding AG: Board Committees (website) · 3 November 2026 · primary source
- F. Hoffmann-La Roche Ltd, Group Legal Compliance: Roche Group Code of Conduct (© 2025) · 3 November 2026 · secondary source
- Roche Holding AG: Risk and opportunity management (website) · 3 November 2026 · secondary source
- Built In London (Roche job advertisement): Senior Procurement AI Governance Manager · 17 August 2026 · secondary source
- KPMG AG: Independent assurance report on selected sustainability information 2025 · 27 January 2026 · primary source
- Roche Holding AG: Roche Annual General Meeting 2026 (media release) · 10 March 2026 · primary source
- F. Hoffmann-La Roche Ltd: Roche Data Ethics Principles · 1 March 2025 · secondary source
2 · Safety
Roche documents approvals, an ISO certification and coordinated vulnerability disclosure, while AI-specific testing is lacking and a recall remains open.
What is documented
- Roche Digital Pathology Dx received FDA 510(k) clearance with an accuracy study; it covers the imaging system, not the AI algorithms1 2
- The Kidney Klinrisk algorithm carries a CE marking under the IVDR and has been validated in a peer-reviewed study by the model developer3 4
- BSI certified Roche Diagnostics International to ISO/IEC 27001:2022 for the navify platform, valid until June 202710
- Roche cites further ISO standards and internal and external penetration testing11 12
- Roche operates a public reporting channel for security researchers and publishes dated advisories, including on AI platform components14
The case against
Recall and vulnerabilities
Sources for Safety (19)
- U.S. Food and Drug Administration: 510(k) Summary K232879 – Roche Digital Pathology Dx (VENTANA DP 200) · 18 June 2024 · primary source
- Roche Diagnostics: Roche receives FDA clearance on its digital pathology solution for diagnostic use (media release) · 18 June 2024 · secondary source
- Roche Holding AG: Roche receives CE mark for AI-based Kidney Klinrisk Algorithm (media release) · 6 October 2025 · secondary source
- Tangri N. et al. (Clinical Kidney Journal, via CiteDrive): Validation of the Klinrisk chronic kidney disease progression model in the FIDELITY population · 1 January 2024 · primary source
- Fierce Biotech: Roche receives FDA breakthrough label for AI-powered lung cancer companion diagnostic test · 3 November 2026 · secondary source
- MedTech Dive: Roche CE mark CGM Accu-Chek SmartGuide · 10 July 2024 · secondary source
- Roche Diagnostics: Image analysis algorithms (digital pathology product page) · 3 November 2026 · secondary source
- U.S. Food and Drug Administration: Artificial Intelligence-Enabled Medical Devices (list) · 4 September 2026 · primary source
- Roche Holding AG: Annual Report 2025 (English) · 29 January 2026 · secondary source
- BSI Assurance UK Limited: ISO/IEC 27001:2022 certificate IS 806759 for Roche Diagnostics International Ltd · 20 September 2024 · primary source
- Roche Diagnostics (navify): Digital trust – Certifications (website) · 3 November 2026 · secondary source
- Roche Diagnostics (navify): Digital trust – Penetration testing (website) · 3 November 2026 · secondary source
- Roche Holding AG: Genentech and NVIDIA enter into strategic AI research collaboration (Investor Update) · 21 November 2023 · secondary source
- Roche Diagnostics: Product Security Advisory (website) · 30 June 2026 · primary source
- U.S. Food and Drug Administration: Class 2 Device Recall navify Digital Pathology (Z-2913-2026) · 7 August 2026 · primary source
- F. Hoffmann-La Roche Ltd: Roche Data Ethics Principles · 1 March 2025 · secondary source
- F. Hoffmann-La Roche Ltd: Roche Artificial Intelligence (AI) Ethics Principles · 1 April 2025 · secondary source
- Abington Cole + Ellery (law firm): Genentech (Cencora / Lash Group) Data Breach · 22 May 2024 · secondary source
- Board Cybersecurity (Incident Tracker): Genentech cybersecurity incident · 6 April 2026 · secondary source
3 · Transparency
Product pages label AI features and regulatory status, but model or system documentation and training data for Roche's own algorithms were not found.
What is documented
- The AI principles commit to disclosing AI-enabled features and the role of humans and to documenting assumptions, limitations and data sources1
- Product pages label Klinrisk as ML-based, SmartGuide as using AI-based predictions and state the regulatory status of the pathology algorithms2 3 4
- The pathology algorithms are described as aids; reviewing the results remains the responsibility of the pathologist2
- The data ethics principles call for information on the origin and quality of data6
The case against
Missing model information
- Model or system cards, information on training data and transparency reports on Roche's own algorithms were not found
- The only peer-reviewed performance study concerns the Klinrisk model, which KlinRisk Inc. developed and Roche only distributes5
Limits not described
- For Klinrisk the target group and purpose are stated, but no limitations3
- Performance limits, error rates and explanations of the predictions of SmartGuide and TROP2 are not documented
Sources for Transparency (6)
- F. Hoffmann-La Roche Ltd: Roche Artificial Intelligence (AI) Ethics Principles · 1 April 2025 · secondary source
- Roche Diagnostics: Image analysis algorithms (digital pathology product page) · 3 November 2026 · secondary source
- Roche Holding AG: Roche receives CE mark for AI-based Kidney Klinrisk Algorithm (media release) · 6 October 2025 · secondary source
- MedTech Dive: Roche CE mark CGM Accu-Chek SmartGuide · 10 July 2024 · secondary source
- Tangri N. et al. (Clinical Kidney Journal, via CiteDrive): Validation of the Klinrisk chronic kidney disease progression model in the FIDELITY population · 1 January 2024 · primary source
- F. Hoffmann-La Roche Ltd: Roche Data Ethics Principles · 1 March 2025 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: within the 36-month window no judgment, regulatory decision, fine or pattern of serious incidents involving Roche's AI systems was found. The FDA recalls concern software without an AI link.
5 · Controversies (2)
- minorVentana Medical Systems initiated a recall of the navify Digital Pathology 2.5 software, classified by the FDA as class 2, because images or metadata in multi-sample cases could be assigned to the wrong sample and slide Seals. 190 units worldwide are affected; no AI link is mentioned. · 20 April 2026 · open
- minorAccording to a law firm notice, a data incident involving medical data of more than 5,800 people was reported to the Texas Attorney General; it is attributed to Genentech and is probably part of the incident at the service provider Cencora/Lash Group. No fine or lawsuit was found. · 22 May 2024 · unklar
6 · What remains open
- Board oversight of AI and cybersecurity; the corporate governance section was not readable
- AI policy with risk classification and approval procedures for own and purchased AI
- Validation, monitoring and performance data for Roche's own AI algorithms
- Regulatory status of the TROP2 companion diagnostic and completion status of the PathAI acquisition at the reference date
- Providers and controls of the generative AI used internally
- Cause, extent and attribution of the 2024 Genentech data incident
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 42 fetches.
Primary documents read (12)
- Roche Annual Report 2025 (English, up to approx. p. 67) and Geschäftsbericht 2025 (German, pp. 1–43)
- Roche Finance Report 2025 (up to p. 41)
- Roche AI Ethics Principles (version 2025-04-01)
- Roche Data Ethics Principles (version March 2025)
- Roche Group Code of Conduct (© 2025, complete with read-more section)
- Roche Board Committees and Risk and opportunity management (web pages)
- KPMG Independent assurance report 2025
- FDA 510(k) Summary K232879
- FDA Recalls Z-2913-2026 and Z-1476-2026
- BSI-Zertifikat ISO/IEC 27001:2022 IS 806759
- Roche Diagnostics Product Security Advisory
- Roche press releases on Klinrisk, PathAI, 2026 Annual General Meeting, Genentech–NVIDIA, and FDA clearance for digital pathology
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0