Principles on data and AI ethics, but little evidence of oversight and testing
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
AstraZeneca describes an AI governance framework and an AI risk process, but both are documented only as self-description without verifiable procedural documentation.
What is documented
- The principles provide for governance proportional to impact and risk and for a review of the data protection standards of third-party AI providers8
- The Chief Data Officer describes an AI risk process in the Enterprise Data Office and a combined review of data protection, security and AI5 9
- At management level, the Chief Digital Officer, the Chief Data Officer and the role of Director of Data and AI Policy are named4 5 6
- More than 50,000 employees took part in the “Thriving in the Age of AI” programme in 2025; more than 17,000 are AI-certified7 14
- The AZethics reporting channels are operated by an independent third party15 13
The case against
Board-level oversight not documented
Self-description without audit
Sources for Responsibility (18)
- AstraZeneca PLC: Corporate governance (web page) · 3 November 2026 · secondary source
- AstraZeneca PLC: Science Committee – Terms of Reference · 1 June 2025 · secondary source
- AstraZeneca PLC: Sustainability Committee – Terms of Reference · 1 January 2026 · secondary source
- AstraZeneca: How we're upskilling and preparing our workforce for the age of AI · 11 April 2025 · secondary source
- CDO Magazine: Should AI Governance Stand on Its Own? AstraZeneca CDO Weighs In · 13 November 2025 · secondary source
- Pink Sheet (Citeline): AstraZeneca: AI Regulations 'Tricky' To Interpret & Navigate · 15 November 2024 · secondary source
- AstraZeneca PLC / U.S. SEC (EDGAR): Annual Report and Form 20-F Information 2025 (Exhibit 15.1) · 24 February 2026 · secondary source
- AstraZeneca: Data and AI ethics (web page) · 1 January 2026 · secondary source
- CDO Magazine: Inside AstraZeneca's AI Strategy: CDO Brian Dummann on Innovation, Governance, and Speed · 29 October 2025 · secondary source
- AstraZeneca PLC / U.S. SEC (EDGAR): Form 20-F for the fiscal year ended December 31, 2025 (main document) · 24 February 2026 · secondary source
- Mökander, Sheth, Gersbro-Sundler, Blomgren (AstraZeneca), Floridi, arXiv / Frontiers in Computer Science: Challenges and Best Practices in Corporate AI Governance: Lessons from the Biopharmaceutical Industry Wolfstre
- AstraZeneca: AstraZeneca Standard – Data Privacy (STND-0002094, Version 6.0) · 1 January 2023 · secondary source
- AstraZeneca: Sustainability Impact Publication 2026 · 1 April 2026 · secondary source
- Fortune (Sheryl Estrada): How AstraZeneca's 17,000 AI-certified employees are helping it reach a 'stretch goal' of $80 billion in revenue · 30 April 2026 · secondary source
- AstraZeneca: Code of Ethics (English) · 3 November 2026 · secondary source
- Jakob Mökander, Luciano Floridi (Oxford Internet Institute), arXiv / AI and Ethics: Operationalising AI governance through ethics-based auditing: An industry case study · 31 December 2021 · secondary source
- AstraZeneca: Sustainability Data Annex 2025 · 3 November 2026 · primary source
- AstraZeneca PLC: Annual Report and Form 20-F Information 2025 (PDF) · 24 February 2026 · secondary source
2 · Safety
Data and system protection is documented with figures and a certificate for Evinova, whereas validation and monitoring of its own AI systems are not.
What is documented
- The 2025 data annex reports zero material cyber incidents and zero material breaches of personal data for 2024 and 20254
- Evinova's information security management system is certified to ISO/IEC 27001:2022 by BSI6
- The data privacy standard governs risk assessment, third-party due diligence and incident reporting7
- The Form 20-F names unreliability and malfunction as risks of AI3
The case against
Missing evidence of AI testing
- There is no information on validation, testing, red-teaming and monitoring of QCS, the Evinova agents, the Modella models and AZerca
- The TROP2 companion diagnostic using the QCS platform has an FDA Breakthrough Designation, but no documented approval1
- According to the company, the figures were reviewed by KPMG; the separate KPMG report on the data annex does not cover them4 5
Incidents and proceedings
- According to the indictment by the Shenzhen prosecutor, the Chinese subsidiary allegedly collected patient data unlawfully; no AI link is documented8 9
- Lapsus$ claims to have stolen around 3 GB of internal data; AstraZeneca has neither confirmed nor commented on this10 11
- No public channel for reporting vulnerabilities was found
Sources for Safety (11)
- Roche (via BioSpace): Roche granted FDA Breakthrough Device Designation for first AI-driven companion diagnostic for non-small cell lung cancer · 29 April 2025 · secondary source
- Evinova / Business Wire: Evinova: strategic collaboration with Astellas, AstraZeneca and Bristol Myers Squibb for the AI-native platform · 18 February 2026 · secondary source
- AstraZeneca PLC / U.S. SEC (EDGAR): Form 20-F for the fiscal year ended December 31, 2025 (main document) · 24 February 2026 · secondary source
- AstraZeneca: Sustainability Data Annex 2025 · 3 November 2026 · primary source
- KPMG LLP: Independent limited assurance report – Sustainability Data Annex 2025 · 8 April 2026 · primary source
- Evinova (AstraZeneca Group): Evinova Secures ISO 27001 Certification, Elevating Data Security Standards · 18 March 2026 · primary source
- AstraZeneca: AstraZeneca Standard – Data Privacy (STND-0002094, Version 6.0) · 1 January 2023 · secondary source
- Fierce Pharma: China indicts AstraZeneca and former exec Leon Wang over data, trade charges · 1 November 2025 · secondary source
- Pharmaceutical Executive: Former AstraZeneca executive charged with several offenses in China · 12 February 2026 · secondary source
- SecurityWeek: Extortion Group Claims It Hacked AstraZeneca · 26 March 2026 · secondary source
- Bitdefender (Hot for Security): Lapsus$ claims AstraZeneca breach exposes code and credentials · 26 March 2026 · secondary source
3 · Transparency
AstraZeneca promises to inform people when they interact with AI, but documents neither labelling nor model or system information on its own tools.
What is documented
- The principles commit to being open about the benefits, strengths and limitations of AI and to disclosing interactions with AI1
- The Form 20-F names bias, unreliability and malfunction as risks of AI4
- Annual and sustainability reports describe uses of AI in research, screening and follow-up care2 3
- According to the company, more than 40 per cent of its genomic data comes from under-represented population groups5
The case against
Missing disclosure
- No model or system cards were found for QCS, the Evinova agents and the Modella models
- Labelling towards patients for AZerca and screening applications is not documented
- There is no stand-alone responsible AI or transparency report8
- Information on the limitations and error rates of individual products is missing
Sources for Transparency (8)
- AstraZeneca: Data and AI ethics (web page) · 1 January 2026 · secondary source
- AstraZeneca PLC / U.S. SEC (EDGAR): Annual Report and Form 20-F Information 2025 (Exhibit 15.1) · 24 February 2026 · secondary source
- AstraZeneca: Sustainability Impact Publication 2026 · 1 April 2026 · secondary source
- AstraZeneca PLC / U.S. SEC (EDGAR): Form 20-F for the fiscal year ended December 31, 2025 (main document) · 24 February 2026 · secondary source
- AstraZeneca PLC: Sustainability Investor Call July 2026 (presentation) · 6 July 2026 · secondary source
- Tissue Pathology (reproducing the Modella AI announcement): Modella AI Announces Acquisition by AstraZeneca to Advance AI-Driven Oncology R&D at Global Scale · 14 January 2026 · secondary source
- Evinova / Business Wire: Evinova: strategic collaboration with Astellas, AstraZeneca and Bristol Myers Squibb for the AI-native platform · 18 February 2026 · secondary source
- AstraZeneca: Sustainability resources (web page) · 3 November 2026 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: in the research period there is no judgment, no authority decision and no fine concerning AI systems, and the ongoing indictment in China concerns data protection without an AI link.
5 · Controversies (2)
- mediumThe Shenzhen prosecutor indicted the Chinese subsidiary over the unlawful collection of patient data and illegal trading; the proceedings are ongoing. · 12 February 2026 · open
- minorThe Lapsus$ group claims to have stolen around 3 GB of internal data; AstraZeneca has neither confirmed nor commented on this. · 26 March 2026 · unklar
6 · What remains open
- Board or committee oversight of AI and cybersecurity; the governance chapter of the 2025 annual report was not readable
- Current status and auditing of AI governance since 2022, such as inventory, approvals and audits
- Validation and monitoring of its own AI systems such as QCS, the Evinova agents and the Modella models
- Labelling of AI towards patients
- Primary documents and outcome of the criminal proceedings against the Chinese subsidiary
7 · How it was assessed
Cut-off date: 3 November 2026 · 40 searches, 40 fetches.
Primary documents read (8)
- Form 20-F 2025 (main document, Item 3.D) and Exhibit 15.1 (pp. 1–17)
- Annual Report 2025 PDF and accessible version (pp. 1–24)
- Form 20-F 2024 (Item 3.D) and Annual Report 2024 (pp. 1–24)
- Data and AI ethics (webpage, January 2026)
- Terms of Reference Science Committee (June 2025) and Sustainability Committee (January 2026)
- Code of Ethics; Data Privacy Standard v6.0
- Sustainability Data Annex 2025; KPMG assurance report on the Data Annex 2025; Sustainability Impact Publication 2026; Sustainability Investor Call July 2026
- Evinova announcements on ISO 27001 certification and the AI platform
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0