Concrete security organisation, but AI rules only since 2026 and without evidence of implementation
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
The board, the Audit and Risk Committee and reporting channels are documented, but for AI there is only the policy published in July 2026 and a task force without evidence of implementation.
What is documented
- The board bears overall responsibility for risk governance; several risk bodies work under the Audit and Risk Committee following the three-lines-of-defence model5 3
- 465 reports were received through all channels in 2025; anonymous reports are possible and retaliation is prohibited3
- The Responsible AI Policy of July 2026 assigns AI governance to the board, the Audit & Risk Committee, the Risk Management Steering Committee and an AI Center of Excellence1 2
- Before an AI system is deployed, responsible persons must be named and possible harms assessed; manipulative AI, social scoring and unauthorised biometrics are excluded1 2
- The 20-F names AI risks and reports an “AI risk mitigation taskforce” based on industry frameworks6
The case against
Missing evidence of implementation
- The policy names no responsible persons, no approving body, no complaint channels and no fixed review cycle1
- The mandate, leadership and results of the AI Center of Excellence and the AI task force are not documented
- Risk classification, approval procedures for in-house and purchased AI, AI training figures and an external audit are missing
Sources for Responsibility (8)
- TSMC: Responsible Artificial Intelligence Policy (English) · July 2026 · secondary source
- TSMC: 負責任人工智慧政策 (Chinese version) · July 2026 · secondary source
- TSMC: 2025 Annual Report, Chapter 3 Corporate Governance (English) · 3 November 2026 · primary source
- TSMC: 2025 Annual Report, Chapter 5 (Operations, incl. 5.3 Smart Manufacturing, 5.4/5.5 Information Security) (English) · 3 November 2026 · primary source
- TSMC: 2025 Annual Report, Chapter 6 (6.2 Risk Management) (English) · 3 November 2026 · primary source
- TSMC (SEC EDGAR): Form 20-F FY2025, XBRL page R46 “Cybersecurity Risk Management and Strategy Disclosure” · 16 April 2026 · primary source
- TSMC: 2025 TSMC Sustainability Report (English) · 3 November 2026 · primary source
- TSMC ESG: Announcement of the 2025 Sustainability Report · 3 November 2026 · secondary source
2 · Safety
TSMC documents data and system protection concretely and uncovered a trade secret leak itself, but a US investigation into TSMC dies in Huawei chips is under way, and tests of its internal AI are not described.
What is documented
- A CISO has been responsible since 2025; two committees chaired by him report to the Audit and Risk Committee every six months2
- ISO 27001 certification, more than NT$1 billion in security spending in 2025 and, according to the annual report, no material security incidents in 20252
- The 20-F describes penetration tests, red-team and ransomware exercises as well as supplier assessments and on-site audits3
- The SEMI E187 standard initiated by TSMC has been a requirement for procuring new equipment since 20237
- Own monitoring detected the leak of 2-nm secrets in 2025; a court sentenced the perpetrators to prison terms5 4
- In October 2024 TSMC itself reported that a customer chip may have reached a restricted entity, and it is cooperating with the authorities10 14
The case against
Export controls
Gaps in AI use
- Validation, tests and monitoring of the AI models in manufacturing and engineering are not described
- The AI-assisted monitoring of communication with employees is disclosed only in the privacy policy6
Sources for Safety (14)
- TSMC: 2025 TSMC Sustainability Report (English) · 3 November 2026 · primary source
- TSMC: 2025 Annual Report, Chapter 5 (Operations, incl. 5.3 Smart Manufacturing, 5.4/5.5 Information Security) (English) · 3 November 2026 · primary source
- TSMC (SEC EDGAR): Form 20-F FY2025, XBRL page R46 “Cybersecurity Risk Management and Strategy Disclosure” · 16 April 2026 · primary source
- Al Jazeera: Taiwan court hands down jail terms in TSMC trade secrets case · 27 April 2026 · secondary source
- Neowin: TSMC's trade secret meltdown exposed by internal monitoring · August 2025 · secondary source
- TSMC: Privacy Policy (last updated 2026-03-24) · 24 March 2026 · primary source
- TSMC ESG: A Responsible Purchaser · 3 November 2026 · primary source
- Bloomberg Línea: «AI is like a three-year-old Superman»: TSMC executive explains his caution with this technology · 16 September 2026 · secondary source
- TSMC: 2025 Annual Report, Chapter 3 Corporate Governance (English) · 3 November 2026 · primary source
- TSMC: 2025 Annual Report, Chapter 6 (6.2 Risk Management) (English) · 3 November 2026 · primary source
- TSMC (SEC): Annual Report on Form 20-F for the fiscal year 2025 (PDF) · 16 April 2026 · primary source
- Reuters (Karen Freifeld), via Claims Journal: TSMC Could Face $1B or More Fine From US Probe, Sources Say · 8 April 2025 · secondary source
- Bloomberg News (Mackenzie Hawkins), via Bloomberg Law: Huawei Used TSMC, Samsung, SK Hynix Components in Top AI Chips · 3 October 2025 · secondary source
- Silicon UK (after Reuters/Financial Times): TSMC Alerts US Of Attempt To Circumvent China AI Chip Restriction · 22 October 2024 · secondary source
3 · Transparency
Model cards do not apply as there are no in-house AI models; AI is disclosed only in the 20-F risk factors, in the privacy policy and as a principle of the policy.
What is documented
- The 20-F names AI risks such as data protection, security gaps, bias, lack of transparency and unexpected behaviour3 2
- The privacy policy discloses that communication with employees is monitored automatically and with AI to prevent data loss4
- The policy requires informing people about interaction with AI and calls for interpretable results and human decision-making1 5
The case against
Missing disclosure
- There is no transparency or responsible AI report and no overview of the AI systems in use and their purposes
- Limits and human control points of AI in manufacturing, security and HR processes are not documented
- There is no information on the data sources of the internal AI systems
Sources for Transparency (8)
- TSMC: Responsible Artificial Intelligence Policy (English) · July 2026 · secondary source
- TSMC (SEC): Annual Report on Form 20-F for the fiscal year 2025 (PDF) · 16 April 2026 · primary source
- TSMC (SEC EDGAR): Form 20-F FY2025, XBRL page R46 “Cybersecurity Risk Management and Strategy Disclosure” · 16 April 2026 · primary source
- TSMC: Privacy Policy (last updated 2026-03-24) · 24 March 2026 · primary source
- TSMC: 負責任人工智慧政策 (Chinese version) · July 2026 · secondary source
- Bloomberg Línea: «AI is like a three-year-old Superman»: TSMC executive explains his caution with this technology · 16 September 2026 · secondary source
- TSMC: 2025 Annual Report, Chapter 5 (Operations, incl. 5.3 Smart Manufacturing, 5.4/5.5 Information Security) (English) · 3 November 2026 · primary source
- TSMC: 2025 Annual Report, Chapter 6 (6.2 Risk Management) (English) · 3 November 2026 · primary source
4 · Why the harm rule does not apply
The harm rule does not apply: there is no ruling, no regulatory decision and no fine concerning own AI systems. The US investigation is pending and concerns manufacturing for third parties.
5 · Controversies (2)
- mediumAccording to media reports, the US Department of Commerce is investigating manufacturing for Sophgo after TSMC dies were found in Huawei AI processors; TSMC had reported the case itself. · 8 April 2025 · open
- minorA Taiwanese court sentenced former TSMC employees to prison for leaking 2-nm trade secrets; TSMC is the injured party. · 27 April 2026 · abgeschlossen
6 · What remains open
- How the Responsible AI Policy is implemented: mandate and leadership of the AI Center of Excellence, work and results of the AI task force, risk framework used.
- Rules for internal and purchased generative AI tools in handling confidential process and customer data.
- Tests, validation and monitoring of the AI models in manufacturing, quality and security.
- Outcome of the US export control investigation and concrete improvements to customer checks for AI chips.
- Scope of the AI-assisted communication monitoring and information provided to those affected.
- The Corporate Governance and Information Security chapters of the 2025 Sustainability Report and the Information Security Statement were not read.
7 · How it was assessed
Cut-off date: 3 November 2026 · 41 searches, 36 fetches.
Primary documents read (12)
- Responsible Artificial Intelligence Policy, English and Chinese (July 2026)
- 2025 TSMC Sustainability Report, English (to p. 37, two retrievals)
- 2025 Annual Report, Chapter 3 Corporate Governance (complete, two retrievals)
- 2025 Annual Report, Chapter 5 (to approx. p. 117)
- 2025 Annual Report, Chapter 6 Risk Management (complete, two retrievals)
- 2025 Annual Report, Chapter 7 Sustainability (to approx. p. 170)
- Form 20-F FY2025 (PDF to p. 19), XBRL R46 Cybersecurity, R42 Contingent Liabilities, Filing Index and FilingSummary
- Form 6-K Media Statement dated 2025-11-25
- Privacy Policy (2026-03-24)
- ESG page «A Responsible Purchaser» (SEMI E187)
- Resolutions of the 2025 Annual General Meeting
- NBIM exclusion list (extract)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0