AI bodies and open security reporting, but several flaws in chips and AI software
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Intel describes committee oversight, a Responsible AI team, a RAI Council and AI assessments for consequential applications, but figures, external audits and an explicit AI mandate in the proxy statement are missing.
What is documented
- According to the CR report, AI governance is anchored with the Audit & Finance Committee and delegated to an Ethics and Compliance Oversight Committee1
- There is a Responsible AI team under the CTO and a RAI Council; consequential use cases undergo Responsible AI Assessments1
- The Audit Committee oversees product security and cyber risks; the full board oversees AI strategy2 3
- Principles and an ethical impact assessment provide for multidisciplinary reviews across the life cycle6 7 8
- The human rights principles provide for an externally run Integrity Line and restricting business in cases of misuse5 4
The case against
Self-description without figures
- The mandate of the RAI Council and the number and results of Responsible AI Assessments are not published
- No AI figures, AI certifications or external audit of AI governance were found9
- Rules for purchased and internally used AI and a risk classification are missing
Sources for Responsibility (9)
- Intel Corporation: Corporate Responsibility Report 2025-26 – Governance, Management and Oversight · 3 November 2026 · primary source
- Intel Corporation: Proxy Statement 2026 (accessible PDF version) · 13 May 2026 · primary source
- Intel Corporation / U.S. SEC: Proxy Statement DEF 14A 2026 · 13 May 2026 · primary source
- Intel Corporation: Corporate Responsibility Report 2025-26 (Full Report) · 17 June 2026 · secondary source
- Intel Corporation: Global Human Rights Principles · 1 June 2025 · secondary source
- Intel Corporation: Responsible AI Principles (PDF) · 3 November 2026 · secondary source
- Intel Corporation (Lama Nachman, Intel Labs): How Intel is Refining Its Approach to Responsible AI · 2 April 2024 · secondary source
- Intel Corporation: Responsible AI · 13 October 2026 · secondary source
- Intel Corporation: Corporate Responsibility Report 2025-26 – GRI Index · 3 November 2026 · secondary source
2 · Safety
Intel runs a documented product security programme with public advisories, including for AI software, but a critical flaw in its AI software, circumventable SGX protection and the Downfall lawsuit weigh against this.
What is documented
- According to the Product Security Report, Intel fixed 374 vulnerabilities in 2024, about 96 % found internally according to the company1 2
- Advisories on Neural Compressor, OpenVINO Model Server and AI Playground state CVSS scores and fixed versions4 5 6
- The Canadian Centre for Cyber Security officially relays Intel advisories7
- The board oversees cyber and product security; the CISO reports at least twice a year8
- Vulnerabilities can be reported via Intigriti and the PSIRT12
- Intel Labs develops robustness tests for generative models3
The case against
Flaws in chips and software
- Vulnerability CVE-2024-22476 in Neural Compressor allowed remote code execution without authorisation; further flaws followed in OpenVINO Model Server and AI Playground4 14 5 6
- With TEE.fail, SGX memory protection can be bypassed with physical access; Intel classifies this as outside the protection scope9
- Via flaws in internal portals, data on more than 270,000 employees could be accessed10
Sources for Safety (15)
- Intel Corporation (IPAS, Jerry Bryant): 2024 Intel Product Security Report Released · 10 February 2025 · primary source
- SecurityWeek: Intel Patched 374 Vulnerabilities in 2024 · 10 February 2025 · secondary source
- Intel Labs: Responsible AI Research · 3 November 2026 · secondary source
- Intel Corporation: Intel Neural Compressor Software Advisory INTEL-SA-01109 · 14 May 2024 · primary source
- Intel Corporation: OpenVINO Model Server Software Advisory INTEL-SA-01272 · 13 May 2025 · primary source
- Intel Corporation: AI Playground Software Advisory INTEL-SA-01352 · 12 August 2025 · primary source
- Canadian Centre for Cyber Security: Intel security advisory (AV26-453) · 12 May 2026 · primary source
- Intel Corporation: Proxy Statement 2026 (accessible PDF version) · 13 May 2026 · primary source
- Intel Corporation: Intel Security Announcement INTEL-2025-10-28-001 – TEE.fail · 28 October 2025 · primary source
- The Register: Intel website flaws exposed data on 270,000 employees · 20 August 2025 · secondary source
- Intel Corporation: Global Human Rights Principles · 1 June 2025 · secondary source
- Intel Corporation: Intel Security Bulletins · 2 November 2026 · primary source
- Tom's Hardware: Intel suspends bug bounty programme that paid up to $100,000 per flaw — new Intigriti disclosure programme offers no rewards · 19 September 2026 · secondary source
- Dark Reading: Intel Discloses Max Severity Bug in Its AI Model Compression Software · 17 May 2024 · secondary source
- Intel Corporation / intc.com: Form 10-Q Q2 2026, note Contingencies · 27 June 2026 · primary source
3 · Transparency
Intel discloses security vulnerabilities in detail and publishes a model card for one of its own models, but it dates from 2023, and an AI transparency report and labelling of AI-generated content are missing.
What is documented
- The Neural-Chat model card states the training dataset, evaluation results and limitations such as false or biased outputs7
- Public advisories and bulletins disclose vulnerabilities with CVSS scores and affected versions1 2 3 4
- The 10-K names AI market and product security risks5
- Intel Labs offers tools for model explainability and interpretation6 8
The case against
Narrow model information
Missing reports and labelling
- No AI transparency report was found
- No labelling of AI-generated content, for example in AI Playground, is documented; C2PA membership is only a company statement6
Sources for Transparency (10)
- Intel Corporation: Intel Neural Compressor Software Advisory INTEL-SA-01109 · 14 May 2024 · primary source
- Intel Corporation: OpenVINO Model Server Software Advisory INTEL-SA-01272 · 13 May 2025 · primary source
- Intel Corporation: AI Playground Software Advisory INTEL-SA-01352 · 12 August 2025 · primary source
- Intel Corporation: Intel Security Bulletins · 2 November 2026 · primary source
- Intel Corporation / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-27 · 22 January 2026 · primary source
- Intel Corporation: Responsible AI · 13 October 2026 · secondary source
- Intel / Hugging Face: Model Details: Neural-Chat-v3 (model card) · 1 October 2023 · primary source
- Intel Labs: Responsible AI Research · 3 November 2026 · secondary source
- Intel Corporation: Responsible AI Principles (website) · 18 June 2025 · secondary source
- Intel OpenVINO / Hugging Face: OpenVINO/Phi-3.5-mini-instruct-fp16-ov (model card) · 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, no authority decision and no fine concerning harm from AI use. The remaining cases are pending or vulnerabilities without documented harm.
5 · Controversies (5)
- mediumWith TEE.fail, researchers showed that SGX memory protection on Xeon platforms can be bypassed with an interposer; Intel names no specific remedy. · 28 October 2025 · unklar
- mediumIntel disclosed a critical flaw in its Neural Compressor AI software that allowed remote code execution; further flaws followed in OpenVINO Model Server and AI Playground. · 14 May 2024 · unklar
- mediumA class action over the Downfall processor flaw accuses Intel of having known about the flaw; the case is pending after partial dismissal. · 1 November 2023 · open
- minorA researcher was able to access data on more than 270,000 employees via flaws in four internal Intel portals; Intel says it fixed the flaws promptly. · 18 August 2025 · unklar
- minorThe state-affiliated Cybersecurity Association of China called for a security review of Intel products; a formal review is not documented. · 16 October 2024 · unklar
6 · What remains open
- Mandate and working methods of the RAI Council and the number of Responsible AI Assessments carried out
- AI-related risk factors and Item 1C in the 10-K
- Status of remedies for Downfall and the attacks on SGX and TDX
- Reasons for suspending the bug bounty rewards
- Labelling of AI-generated content in its own applications
7 · How it was assessed
Cut-off date: 3 November 2026 · 41 searches, 40 fetches.
Primary documents read (9)
- Form 10-K FY2025 (S01, two retrievals; cover page to MD&A, Items 1A/1C/3 not included)
- Form 10-Q Q2 2026, Note Contingencies (S11, complete)
- Proxy Statement DEF 14A 2026 (S08, three retrievals, partial) and accessible PDF version (S09, two retrievals, risk oversight pp. 22–25)
- Form 8-K Voting Results 2026 (S10)
- Corporate Responsibility Report 2025-26 (S03, two retrievals to p. 36), Executive Summary (S04), GRI Index (S30), Governance section (S31)
- Responsible AI Principles PDF and website (S05, S28), Responsible AI page (S07), Intel Labs page (S36), Global Human Rights Principles (S27)
- Security advisories INTEL-SA-01109, -01272, -01352, bulletin TEE.fail and bulletin overview (S20, S34, S40, S17, S37)
- Model cards Neural-Chat-v3 and OpenVINO Phi-3.5 (S22, S35)
- Official and independent findings: press release Court of Justice of the EU 153/25 (S29), Canadian Centre for Cyber Security AV26-453 (S39)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0