Certified AI management, but serious data breaches and little evidence of AI testing
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Oracle documents board oversight, an external ISO 42001 certification and contractual review duties, but names no person responsible for AI and no AI metrics.
What is documented
- According to the 2026 proxy, the full board oversees AI risks and opportunities, including responsible deployment and data governance1
- Schellman certified five AI management systems (OCI AI Services, Oracle Health, Life Sciences, Fusion SaaS, NetSuite) to ISO/IEC 42001:20233 4
- The AI Terms require review before deployment and after material changes, alignment with the NIST AI RMF and an inventory5
- According to its own statement Oracle forgoes systems prohibited under Art. 5 of the EU AI Act and names Oracle AI Review as a mandatory gate6
- An Integrity Helpline and, according to Oracle, a reporting channel for AI risks and incidents exist8 3
The case against
Gaps in accountability
Sources for Responsibility (15)
- Oracle Corporation (SEC EDGAR): Definitive Proxy Statement (DEF 14A) for the annual general meeting of 18 November 2026 · 25 September 2026 · primary source
- Oracle Corporation (SEC EDGAR, XBRL R11): 10-K FY2026 – Cybersecurity Risk Management, Strategy, and Governance (Item 1C) · 3 November 2026 · primary source
- Oracle (Oracle Cloud Infrastructure Blog): Raising the bar for trustworthy AI at Oracle · 16 June 2026 · primary source
- Oracle (Oracle Help Center): What AI governance and standards accreditations does Oracle have for Fusion Applications? · 3 November 2026 · primary source
- Oracle: Oracle AI Terms · 16 September 2026 · primary source
- Oracle: ISO 42001: Principles into practice (German-language country page) · 3 November 2026 · secondary source
- Oracle (press release): Oracle and Rafael to provide cloud-based AI solutions for defence missions · 10 September 2024 · primary source
- Oracle (filed in the German Bundestag lobbying register): 2017 Oracle Code of Ethics and Business Conduct · 3 November 2026 · secondary source
- Oracle: Human Rights Statement (Version 1) · 3 November 2026 · secondary source
- Oracle: Social Impact – Practices · 3 November 2026 · secondary source
- Oracle Corporation (SEC EDGAR): Form 8-K – voting results of the 2025 annual general meeting · 18 November 2025 · primary source
- Oracle Corporation (SEC EDGAR): Form 8-K – voting results of the 2024 annual general meeting · 14 November 2024 · primary source
- PPC Land: Oracle's $115 million privacy settlement: what consumers need to know · 19 July 2024 · secondary source
- Bloomberg Law: Oracle's $115m Deal Affirmed After Ad Tech Arm Shuttered · 17 February 2026 · secondary source
- Oracle Corporation (SEC EDGAR): Form 10-Q for the quarter ended 31 August 2026 · 3 November 2026 · primary source
2 · Safety
Oracle documents an externally certified security programme and protective features for generative AI, but three incidents in its own systems in 2025 and the lack of red-teaming weigh heavily.
What is documented
- Item 1C describes penetration testing, tabletop exercises, supplier risk management and certifications to SOC 1/2, ISO 27001/27017/27018 and CSA STAR5
- A named SVP is responsible for security and reports quarterly; the Finance and Audit Committee provides oversight5
- OCI Generative AI offers content moderation and detection of prompt injection and personal data, with documented limits1
- Prompts are not stored, and customer data is neither passed to model providers nor used to train general models6
- The AI Terms prohibit circumventing safeguards, deceptive deepfakes and child exploitation14
The case against
Incidents in its own systems in 2025
- According to reports, an attacker accessed Oracle Cloud login servers; Oracle initially denied a breach and later acknowledged two «obsolete» servers7 8
- According to a notification to Texas, access to legacy Cerner servers of Oracle Health affected almost 20 million people; the negligence claims were not dismissed9 10 11
- Clop exploited the CVE-2025-61882 vulnerability in E-Business Suite to steal customer data before Oracle provided an emergency patch13
Sources for Safety (15)
- Oracle (Oracle Help Center): Guardrails for OCI Generative AI · 29 May 2026 · primary source
- Oracle Health: Developing trustworthy AI in healthcare (White Paper v2) · 3 November 2026 · secondary source
- Oracle (press release): Oracle Health expands Clinical AI Agent with automated coding, dictation, chart review · 19 August 2026 · secondary source
- Oracle (Oracle Cloud Infrastructure Blog): Raising the bar for trustworthy AI at Oracle · 16 June 2026 · primary source
- Oracle Corporation (SEC EDGAR, XBRL R11): 10-K FY2026 – Cybersecurity Risk Management, Strategy, and Governance (Item 1C) · 3 November 2026 · primary source
- Oracle (Oracle Help Center): Data Handling in Generative AI · 3 November 2026 · primary source
- American Bar Association (Health Law Section): Oracle cloud breaches lead to CISA guidance, lawsuits · 16 April 2025 · secondary source
- CSO Online: Oracle admits breach of 'obsolete servers,' denies main cloud platform affected · 10 April 2025 · secondary source
- Tech Insider: Oracle Health breach exposes nearly 20 million people · 2 October 2026 · secondary source
- Paubox: Texas says Oracle Health breach affected nearly 20 million · 22 October 2026 · secondary source
- Paubox: Judge rules against Oracle Health breach vendor-only liability defence · 22 July 2026 · secondary source
- Rankiteo (Blog): Oracle Health stuck with negligence claim from data breach suit · 3 November 2026 · secondary source
- BleepingComputer: Oracle patches EBS zero-day exploited in Clop data theft attacks · 5 October 2025 · secondary source
- Oracle: Oracle AI Terms · 16 September 2026 · primary source
- TechTarget: Oracle adds xAI Grok models to OCI · 17 June 2025 · secondary source
3 · Transparency
Oracle documents in detail how customer data is used in its AI services and where the guardrails reach their limits, but does not name the language model in Fusion and has no documented labelling of AI content.
What is documented
- The documentation discloses that customer data is neither stored nor used for general models; classic ML models are retrained separately per tenant7 8
- The guardrails documentation names limits such as personal data detection only in English and performance caveats4
- In the 10-K Oracle names bias, discrimination and lack of transparency as risks of its own AI2
- The AI Terms prohibit customers from deceiving people about the use of AI1
The case against
Gaps in disclosure
- A transparency report and system cards per product were not found
- Fusion does not name the base model used; Oracle Health's model cards could not be checked8 6
- Labelling or provenance information for AI outputs is not documented
- Oracle's account of the 2025 cloud breach was publicly criticised as «wordsmithing»3
Sources for Transparency (8)
- Oracle: Oracle AI Terms · 16 September 2026 · primary source
- Oracle Corporation (SEC EDGAR): Form 10-K for the fiscal year ended 31 May 2026 · 3 November 2026 · primary source
- CSO Online: Oracle admits breach of 'obsolete servers,' denies main cloud platform affected · 10 April 2025 · secondary source
- Oracle (Oracle Help Center): Guardrails for OCI Generative AI · 29 May 2026 · primary source
- Oracle (press release): Oracle Health expands Clinical AI Agent with automated coding, dictation, chart review · 19 August 2026 · secondary source
- Oracle Health: Developing trustworthy AI in healthcare (White Paper v2) · 3 November 2026 · secondary source
- Oracle (Oracle Help Center): Data Handling in Generative AI · 3 November 2026 · primary source
- Oracle (Oracle Help Center): How's user behavior, input, and feedback incorporated into the model? (Fusion AI) · 3 November 2026 · primary source
4 · Why the harm rule does not apply
The harm rule does not apply: within the window there is no judgment, authority decision or fine on harm from Oracle's AI; the 2025 incidents concern the cybersecurity of legacy and login systems.
5 · Controversies (8)
- mediumIn The Privacy Collective's Dutch class action over the Data Management Platform, the Hoge Raad referred the case back to the appeal court in 2026; proceedings continue. · 17 July 2026 · open
- mediumIn 2025 the Clop group exploited the CVE-2025-61882 vulnerability in Oracle E-Business Suite to steal customer data; Oracle released an emergency patch in October. · 5 October 2025 · unklar
- mediumAn attacker claimed in 2025 to have stolen around six million records from Oracle Cloud login servers; Oracle later acknowledged two «obsolete» servers, a class action is ongoing. · 21 March 2025 · open
- mediumIn 2025 unauthorised parties used stolen credentials to access legacy Cerner servers of Oracle Health; according to Texas almost 20 million people are affected, the lawsuit is ongoing. · 20 February 2025 · open
- mediumIn 2024 Oracle America settled the Katz-Lacabe class action over personal profiles created without consent for USD 115 million without admission of liability. · 19 July 2024 · Vergleich
- minorIn 2026 media reported, unconfirmed, that Oracle used an algorithmic selection tool in mass layoffs; no proceedings are documented. · 16 April 2026 · unklar
Further controversies (2)
- minorAfter TikTok USA was taken over by a joint venture with a 15% Oracle stake, suppression of ICE-critical content was alleged in 2026; California is reviewing. · 25 January 2026 · open
- minorIn 2024 Oracle announced with Rafael that it would provide AI systems for defence missions on OCI, including one with automatic target selection; no safeguards were named. · 10 September 2024 · unklar
6 · What remains open
- Independent review of the effectiveness of Oracle AI Review and the guardrails
- Oracle's root cause and remedy reports on the 2025 security incidents
- Usage restrictions for military and security customers
- Base models of Fusion AI and scope of Oracle Health's model cards
- Complaints, authority notifications and the Hoge Raad ruling in the original
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 41 fetches.
Primary documents read (9)
- Form 10-K FY2026 (S01 with read more, XBRL R11 Item 1C S02, XBRL R26 Legal Proceedings S03, filing index and FilingSummary)
- Proxy Statement DEF 14A 2026 (S04, complete with read more)
- Form 8-K voting results 2024 and 2025 (S05, S06)
- Form 10-Q Q1 FY2027 (S27, Legal Proceedings)
- Oracle AI Terms 2026 (S07)
- ISO 42001 notices and product documentation (S08–S14)
- Oracle Health White Paper (S15)
- Human Rights Statement 2024 (S38)
- Code of Ethics 2017 (S37)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0