Skip to content
GLOBALANCE INTELLIGENCEThe economy ahead.
This Week Index 28/100 Sign in
Money & Investments › GI 100 Index
The Great Value ShiftInvestment ThemesGI 100 IndexGI PortfolioStart Future Check
AI FOOTPRINTsolidASSESSMENT 2026GLOBALANCEINTELLIGENCE
Globalance AI Footprint

Assessment as of November 2026 · next assessment 2027.

Result of the assessment by the Globalance research team under the published methodology. Not a certification or quality label.

How we assess ›

How this rating is builtVideo · 0:52
GI 100 Index · Profile

AMD

USAEnablersAssignment FullData quality Medium

Key messagesolid AI Footprint

AMD supplies chips and open software for artificial intelligence, publishes its own models including training data and fixes reported security flaws through an established disclosure process, but serious flaws in chips and software have occurred repeatedly over the past three years. Principles and a body for responsible AI have been announced, but AMD discloses little about how they are implemented and monitored in practice.

Disclosure: the research was supported by an AI model made by Anthropic. AMD is connected to Anthropic (e.g. as investor, partner, supplier, customer or through people). The assessment was therefore reviewed by people.

Opportunities

  • Safetysolid

    AMD runs a documented PSIRT process with coordinated disclosure, CNA status, FIRST membership and a bug bounty and publishes bulletins, including for AI software (S15, S16, S20); independent bodies confirm that serious flaws such as EntrySign were fixed and that mitigations were provided for Sinkclose (S23, S22).

    AMD runs a documented PSIRT process with coordinated disclosure, CNA status, FIRST membership and a bug bounty and publishes bulletins, including for AI software (S15, S16, S20); independent bodies confirm that serious flaws such as EntrySign were fixed and that mitigations were provided for Sinkclose (S23, S22). For its own models, safety benchmarks and red teaming are documented (S18), and against misuse there is an anti-diversion committee and customer screening (S11). Against this stand repeated serious hardware and firmware flaws, including the extraction of LLM responses from GPU memory and the breaking of the confidential computing environment, delayed fixes and a bounty programme that excludes ROCm and Instinct (S24, S25, S17, S16). The usage policy names neither enforcement nor a reporting channel (S09).

  • Transparencysolid

    AMD publishes its own models fully openly with weights, training configuration, datasets and code; the Instella model card lists datasets for each training stage with token counts and licences as well as limitations and prohibited uses (S19, S18).

    AMD publishes its own models fully openly with weights, training configuration, datasets and code; the Instella model card lists datasets for each training stage with token counts and licences as well as limitations and prohibited uses (S19, S18). The alignment methodology and safety benchmarks are disclosed, and limitations such as jailbreaks are acknowledged (S18). Weaknesses: labelling of AI-generated content is not documented, an AI transparency report is missing, model cards contain no safety figures, and a security bulletin for Instinct lists an unexplained CVE (S19, S21).

Risks

  • Responsibilitylimited

    General oversight structures are documented: the Audit and Finance Committee oversees cyber and compliance risks, the Innovation and Technology Committee the risks of new technologies, and the Chief Risk and Compliance Officer reports quarterly (S06, S13).

    General oversight structures are documented: the Audit and Finance Committee oversees cyber and compliance risks, the Innovation and Technology Committee the risks of new technologies, and the Chief Risk and Compliance Officer reports quarterly (S06, S13). An explicit board responsibility for AI is missing from the 2026 proxy statement; the Responsible AI Council, the AI Use Case Assessment and the review framework named in the human rights document (data and model review, red teaming, vendor assessment) are documented only as principles without a mandate, composition or results (S08, S12, S14). There is no AI report, no AI metrics and no AI certification; the signing of the EU code of practice concerns only the subsidiary Silo AI and has not been officially confirmed (S35, S37).

  • Risk area

    Security of the hardware and firmware on which AI models and confidential data run

  • Risk area

    Diversion of chips into weapons and military applications

  • Risk area

    Export of AI accelerators to China under US licence

  • No serious controversy recorded.

What lies behind the assessment

Research of 3 November 2026
45searches
39pages and documents read
10primary documents
5controversies checked
6open points, named openly
The full research: pillars, evidence, controversies and open points
AI Footprint · Enabler · As of 3 November 2026
AMDOverall rating: solid

AMD publishes its own models including training data and fixes reported flaws through an established disclosure process. Principles and a body for responsible AI have been announced, but AMD discloses little about how they are implemented.

Responsibility

40 %
limitedsolidstrong

Committees oversee cyber and technology risks, but responsibility for AI is missing.

  • +Committees oversee cyber risks and risks of new technologies Source ↗
  • +Code of conduct requires review of internal AI uses before deployment Source ↗
  • –No committee with an explicit AI mandate Source ↗
  • –Responsible AI Council only named, without mandate or composition Source ↗
  • –No AI report, no AI metrics, no AI certification
All sources for this pillar (11)

Safety

40 %
limitedsolidstrong

Reported flaws are fixed through a structured process, but serious flaws keep occurring.

  • +Disclosure process with coordinated disclosure and a bug bounty Source ↗
  • +Fix of the EntrySign flaw independently confirmed Source ↗
  • +In-house models tested with safety benchmarks and red teaming Source ↗
  • –Language model responses extractable from GPU memory (LeftoverLocals) Source ↗
  • –Bug bounty programme does not cover ROCm and Instinct Source ↗
All sources for this pillar (16)

Transparency

20 %
limitedsolidstrong

In-house models are open with data and code, but labelling of generated content is missing.

  • +Open weights, training data and code for in-house models Source ↗
  • +Model card lists datasets with licences and limitations Source ↗
  • –No labelling of AI-generated content documented
  • –Instinct bulletin lists a CVE without explanation Source ↗
All sources for this pillar (6)
What is missing for “strong”

These steps are in the company’s own hands.

1Responsibility
Anchor and document AI oversight

Assign AI oversight to a board committee and publish the mandate, composition and results of the Responsible AI Council.

2Safety
Close flaws faster and fully

Disclose the remediation status of LeftoverLocals and extend the bug bounty programme to ROCm and Instinct.

3Transparency
Label generated content

Label AI-generated content from in-house models and publish an AI transparency report with safety figures.

Context: 101 companies assessed

Distribution of overall ratings, no ranking
strong
none yet
solid
← AMD is one of 35
limited

No company reaches “strong” today. Disclosure on how companies handle AI is still young across the market.

Controversies
5 2 medium, 3 minor

No deduction for controversies.

Largest risk area
Flaws in chips for AI and confidential data

Through the LeftoverLocals flaw, other users of an AMD GPU could reconstruct language model responses of other processes from GPU memory.

How thoroughly assessed
33 Evidence

10 primary documents read, 45 searches.

Read the full reasoning and all sources
AMD · Full reasoning

Open models and an established disclosure process, but repeated serious chip flaws

Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.

1 · Responsibility

limitedWeight 40%

AMD has general oversight structures, but the Responsible AI Council and the AI review procedures are documented only as principles without a mandate or results.

What is documented
  • The Audit and Finance Committee oversees IT and cybersecurity risks and receives an annual report from an independent adviser1
  • The Innovation and Technology Committee is responsible for risk mitigation for products based on new technology1
  • The Chief Risk and Compliance Officer reports quarterly; there is annual mandatory training and the anonymous AMD Aware hotline2 5
  • The code of conduct requires an AI Use Case Assessment, approved tools and an AI Security Standard for internal AI uses5
  • According to an excerpt of the 10-K, the cyber programme is aligned with the NIST AI Risk Management Framework6
The case against
AI governance only as a principle
  • In the 2026 proxy statement no committee has an explicit AI mandate, and no person responsible for AI is named1
  • The Responsible AI Council is only named; its mandate, chair and reporting line are not disclosed4
  • Data and model review, red teaming and vendor assessment are named in the human rights policy only as principles, without results3
Missing accountability
  • There is no AI report, no AI metrics, no AI certification and no dedicated channel for reporting AI misuse7
  • The signing of the EU code of practice concerns only the subsidiary Silo AI and has not been officially confirmed10 11
Sources for Responsibility (11)
  1. AMD / U.S. SEC: Proxy Statement DEF 14A 2026 · 13 May 2026 · primary source
  2. AMD: Ethics and Compliance – Our Approach · 3 November 2026 · primary source
  3. AMD: Human Rights Policy (Updated August 2026) · 3 November 2026 · secondary source
  4. AMD: Responsible AI – Our Commitment · 3 November 2026 · secondary source
  5. AMD: AMD Global Code of Conduct (October 2025) · 3 November 2026 · secondary source
  6. Board Cybersecurity (extract from 10-K): ADVANCED MICRO DEVICES INC 10-K Item 1C. Cybersecurity – 2026-02-03 · 3 February 2026 · secondary source
  7. AMD: AMD Responsible AI Use Policy · 3 November 2026 · secondary source
  8. AMD: 2025–26 Corporate Responsibility Report · 18 August 2026 · primary source
  9. AMD / U.S. SEC: Form 8-K, voting results of the 2025 annual general meeting · 14 May 2025 · primary source
  10. AMD Silo AI: AMD Silo AI Signs the EU AI Code of Practice · 20 August 2025 · secondary source
  11. Freevacy: EU Commission publishes GPAI Code of Practice assessment and signatory list · 1 August 2025 · secondary source

2 · Safety

solidWeight 40%

AMD fixes reported flaws through a documented disclosure process and tests its own models, but repeated serious flaws in chips and firmware stand against this.

What is documented
  • AMD runs a PSIRT with coordinated disclosure, CNA status, FIRST membership and a bug bounty via Intigriti4 5
  • AMD publishes security bulletins, including for AI software, for example on four flaws in Ryzen AI Software and the NPU driver6
  • A fix for EntrySign was shipped to customers; mitigations are available for Sinkclose9 8
  • In-house models are evaluated with ToxiGen, CrowS-Pairs, TruthfulQA, adversarial tests and red teaming1
  • Against diversion there is an anti-diversion committee, customer screening, locking software and cooperation with Conflict Armament Research14
The case against
Repeated serious flaws
  • Via LeftoverLocals, LLM responses could be reconstructed from GPU memory; at disclosure AMD was still assessing mitigations, and their status is unclear7
  • Sinkclose, EntrySign and RMPocalypse affected processor firmware and confidential computing environments; no exploitation is documented8 9 10
  • According to press reports, a flaw in AMD AutoUpdate remained open for up to 124 days; AMD refused the bounty payment11
  • The bug bounty programme does not cover ROCm and Instinct5
Misuse and model release
  • The AI usage policy lists prohibitions but neither enforcement nor a reporting channel13
  • The weights of Instella-3B-Instruct are released without safety promises; the model card contains no safety benchmarks2
Sources for Safety (16)
  1. AMD Developer (E. Barsoum et al.): Building Trust Through Open Innovation · 23 October 2025 · primary source
  2. AMD / Hugging Face: Model card Instella-3B-Instruct · 3 November 2026 · primary source
  3. Board Cybersecurity (extract from 10-K): ADVANCED MICRO DEVICES INC 10-K Item 1C. Cybersecurity – 2026-02-03 · 3 February 2026 · secondary source
  4. AMD: Product Security (PSIRT) · 3 November 2026 · primary source
  5. Intigriti: AMD Bug Bounty Program – Detail · 3 November 2026 · primary source
  6. AMD PSIRT: AMD-SB-7037: AMD Ryzen AI Software Vulnerabilities · 2 April 2025 · primary source
  7. Trail of Bits: LeftoverLocals: Listening to LLM responses through leaked GPU local memory · 16 January 2024 · primary source
  8. CERT-EU: Security Advisory 2024-075: Vulnerabilities in AMD CPUs · 12 August 2024 · primary source
  9. oss-security (Openwall), Google disclosure: AMD Zen microcode signature verification vulnerability (CVE-2024-56161) · 6 March 2025 · primary source
  10. SecurityWeek: RMPocalypse: New Attack Breaks AMD Confidential Computing · 3 November 2026 · secondary source
  11. BGR: AMD Refuses To Pay A $10,000 Bug Bounty For A Flaw It Never Caught · 15 June 2026 · secondary source
  12. Mixed News: AMD's Instinct security bulletin names two CVEs for one flaw, and explains only one · 6 October 2026 · secondary source
  13. AMD: AMD Responsible AI Use Policy · 3 November 2026 · secondary source
  14. AMD: 2025–26 Corporate Responsibility Report · 18 August 2026 · primary source
  15. SecurityWeek: AMD Investigating Breach Claims After Hacker Offers to Sell Data · 19 June 2024 · secondary source
  16. AMD: AMD Global Code of Conduct (October 2025) · 3 November 2026 · secondary source

3 · Transparency

solidWeight 20%

AMD discloses the weights, training data and code of its own models and acknowledges limitations, but documents no labelling of AI-generated content and no AI transparency report.

What is documented
  • For Instella, AMD OLMo, Hummingbird and Zebra-Llama, weights, training data, configurations and code are open6
  • The Instella-3B-Instruct model card lists datasets for each training stage with token counts and licences5
  • The model card contains a section on bias, risks and limitations and excludes areas of use5
  • AMD acknowledges limitations regarding jailbreaks and factual accuracy6
  • The 10-K names risks arising from the responsible use of AI4
The case against
Labelling and reporting
  • No information was found on labelling AI-generated content from the in-house image and video models
  • An AI transparency report is missing
  • The Instella model card contains no safety or bias benchmarks5
Unclear disclosure
  • Bulletin AMD-SB-6033 on the AI library RCCL lists two CVE numbers but explains only one; the fix came about eleven weeks earlier3
  • The model cards for Poro and Viking from Silo AI were not reviewed
Sources for Transparency (6)
  1. AMD: AMD Responsible AI Use Policy · 3 November 2026 · secondary source
  2. AMD PSIRT: AMD-SB-7037: AMD Ryzen AI Software Vulnerabilities · 2 April 2025 · primary source
  3. Mixed News: AMD's Instinct security bulletin names two CVEs for one flaw, and explains only one · 6 October 2026 · secondary source
  4. AMD / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-27 · 4 February 2026 · primary source
  5. AMD / Hugging Face: Model card Instella-3B-Instruct · 3 November 2026 · primary source
  6. AMD Developer (E. Barsoum et al.): Building Trust Through Open Innovation · 23 October 2025 · primary source

4 · Why the harm rule does not apply

The harm rule does not apply: within the window there is no judgment, regulatory decision or fine concerning harm caused by AI. The recorded flaws are vulnerabilities without documented harm.

5 · Controversies (5)

6 · What remains open

  • Board responsibility for AI and composition of the Responsible AI Council
  • Evidence of how the AI Use Case Assessment is implemented
  • Remediation status of the LeftoverLocals flaw
  • Item 1C and Item 3 of the 10-K and the governance chapter of the CR report could not be read in the original
  • Labelling of AI-generated content from in-house generative models
  • Enforcement of the AI usage policy and reporting channel for misuse

7 · How it was assessed

Cut-off date: 3 November 2026 · 45 searches, 39 fetches.

Primary documents read (10)
  • Form 10-K FY2025 (S03, two retrievals, Part 1 through Risk Factors; Item 1C/3 not included)
  • Form 10-K/A FY2025 (S04)
  • Form 8-K Voting Results 2025 (S05)
  • Proxy Statement DEF 14A 2026 (S06, two retrievals, complete)
  • 2025–26 Corporate Responsibility Report (S11, two retrievals through p. 52)
  • Responsible AI page (S08) and Responsible AI Use Policy (S09)
  • Global Code of Conduct Oct. 2025 (p.12), Ethics-and-Compliance page (p.13), Human Rights Policy Aug. 2026 (p.14)
  • PSIRT page (S15), Bug Bounty Program (S16), Bulletin AMD-SB-7037 (S20)
  • Model card Instella-3B-Instruct (S19) and developer contribution to security benchmarks (S18)
  • Official and independent findings: CERT-EU 2024-075 (S22), Google EntrySign disclosure (S23), Trail of Bits LeftoverLocals (S24)

Research of 3 November 2026 under Globalance AI Footprint rulebook version 0

We assess documented practice in dealing with AI; this is not an investment recommendation. Three pillars, weights 40/40/20, harm rule.

Research of 3 November 2026 under Globalance AI Footprint rulebook version 0. Documented observations from public sources, no investment recommendation.

Important Notice. This material is provided for information purposes only and does not constitute an offer or solicitation to buy or sell any financial instrument, nor does it constitute investment advice or a recommendation. The GI 100 Index is a rules-based strategy index of Globalance Bank. Investments involve risk, including the loss of capital invested. The final product documents (including FinSA) shall prevail in all cases.

Your company? Information on inclusion in the GI 100 Index →

What comes next