AIWiki
Malaysia
Back to all articles
Ethics & Policytransparencydisclosurewatermarking

AI Transparency

5 min readUpdated November 2026
AI Transparency
Type
AI governance principle
Also called
Disclosure, provenance, explainability of outputs
EU requirement
AI Act Article 50, in force 2 August 2026
Malaysia requirement
MCMC Risk Mitigation Code, in force 1 June 2026
Penalties
Up to €15M/3% turnover (EU); RM10 million (Malaysia)
Related
AI Watermarking, C2PA, AI Ethics, AI Regulation
AI transparency is the principle that people should be able to tell when they are interacting with an AI system and when content they see, hear or read was generated or materially altered by one — and, in practice, the disclosure duties, technical marking schemes and documentation that make that possible.[1] Since 2026 it has moved from voluntary commitment to enforceable law in several jurisdictions.

History

Transparency entered mainstream AI policy through ethics principles and soft-law instruments — the OECD AI Principles, the EU's Ethics Guidelines for Trustworthy AI and model documentation practices such as model cards. Early technical work on AI watermarking and content provenance (C2PA content credentials) provided the mechanisms; regulation supplied the compulsion. The EU AI Act (Regulation (EU) 2024/1689) set a hard deadline, and Malaysia's MCMC issued a binding code under the Online Safety Act 2025, so that by late 2026 the two most visible regimes on this page were both live.[2][3]

Key Concepts

Transparency obligations split by who they bind:

  • Provider obligations. Systems that interact directly with people must make it clear that the user is dealing with AI (unless that is obvious in context), and generative systems must mark their outputs with effective, reliable, robust and interoperable machine-readable signals so the content can be detected as AI-generated.[1][2]
  • Deployer obligations. Users of such systems must disclose deepfakes — AI-generated or manipulated image, audio or video that would falsely appear authentic — and AI-generated text published to inform the public on matters of public interest. Machine-readable marking alone is not enough: the label must be clear, distinguishable and perceivable at first exposure, for example a visible or audible label.[1][2]
  • Exceptions. Evidently artistic, creative, satirical or fictional works face a reduced disclosure regime; text that has undergone genuine human editorial control by someone with editorial responsibility is exempt; law-enforcement authorisations are excluded.[1][3]
Under the EU AI Act, Article 50 applied from 2 August 2026, with a deferral to 2 December 2026 for the machine-readable marking duty covering generative systems already on the market before that date, and to 2 February 2027 for watermark-detection interoperability.[2][4] Non-compliance can attract fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.[3] The European Commission's finalised guidelines and the voluntary Code of Practice on Transparency of AI-Generated Content introduced standardised EU icons for fully AI-generated and AI-modified content.[3]

Transparency is distinct from — but connected to — explainability (why a model produced an output) and auditability (whether its behaviour can be examined). Disclosure tells you that AI was involved; it does not tell you that the output is correct or fair.

Impact

Transparency changes product design: chatbots must greet users with an AI notice, generated media must carry persistent labelling, and detection mechanisms must be made available — in principle free of charge to regulators, law enforcement, media, fact-checkers and researchers.[3] It also creates a compliance and tooling market around provenance, content credentials and detection. Critics note the limits: labels can be stripped, detection of unmarked content is unreliable, and disclosure duties govern visibility rather than the underlying quality or legality of a system.

>Key Takeaways

  • Transparency means disclosure of AI involvement and marking of generated content, not a guarantee of quality.[1]
  • The EU's Article 50 duties have been enforceable since 2 August 2026, with a narrow watermarking grace period to 2 December 2026.[2][4]
  • Machine-readable marks alone never satisfy the human-facing labelling duty.[1][2]
  • Malaysia's Risk Mitigation Code (in force 1 June 2026) imposes equivalent labelling on platforms, with RM10 million penalties.[5][6]

See Also

🇲🇾Malaysian Context

🇲🇾 Relevance to Malaysia: The MCMC's Risk Mitigation Code under the Online Safety Act 2025 took effect on 1 June 2026. It requires licensed social media and messaging platforms to let users identify manipulated and AI-generated content resembling real people or events, to provide tools and guidance for advertisers and users to disclose AI use, and to apply "prominent labels or markings" to realistic AI-made images, audio and video; advertisers must also be verified against government-issued records. Non-compliance carries financial penalties of up to RM10 million.[5][6]

The driver is measurable: complaints to MCMC over deepfake content rose from 917 in 2024 to 3,612 in 2025 and 7,967 as of 15 June 2026, and MCMC submitted 13,122 content-removal requests to platforms for image manipulation between 1 January and 30 June 2026.[7][8] Political deepfakes have been a particular concern, including fabricated content depicting the King and the Prime Minister promoting investments.[9] MCMC has also studied licensing requirements for AI applications and is equipping itself and CyberSecurity Malaysia with forensic tools to verify whether content is AI-generated, looking at metadata as well as visible labels.[9] For Malaysian businesses, transparency obligations now sit alongside the Malaysia AI Governance Framework, PDPA requirements and sector guidance such as BNM's AI guidelines — making disclosure a routine part of campaign, product and publication workflows.

References

  1. ↑European Commission. (2026). Transparency obligations under Article 50 of the AI Act. https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act
  2. ↑McCann FitzGerald. (2026). AI Transparency: European Commission's Guidelines on Article 50 — Deployer Obligations. https://www.mccannfitzgerald.com/knowledge/data-privacy-and-cyber-risk/ai-transparency-european-commissions-guidelines-on-article-50-part-2-deployer-obligations
  3. ↑Paul, Weiss. (2026). EU Finalises Transparency Rules for AI-Generated Content. https://www.paulweiss.com/insights/client-memos/eu-finalises-transparency-rules-for-ai-generated-content
  4. ↑Cloud Security Alliance. (2026). EU AI Act Article 50 Watermarking Grace Period Ends December 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-eu-ai-act-article50-watermarking-deadline
  5. ↑Malay Mail. (2026). Malaysia's new online safety rules kick in today. https://www.malaymail.com/amp/news/malaysia/2026/06/01/malaysias-new-online-safety-rules-kick-in-today-what-changes-for-social-media-users/221673
  6. ↑The Star. (2026). Election deepfakes now 'dangerously accessible', says cyber consumer group. https://www.thestar.com.my/news/nation/2026/11/13/election-deepfakes-now-dangerously-accessible-says-cyber-consumer-group
  7. ↑New Straits Times. (2026). MCMC triggers purge of 12,000 deepfakes as AI scam complaints surge. https://www.nst.com.my/news/nation/2026/07/1495725/mcmc-triggers-purge-12000-deepfakes-ai-scam-complaints-surge
  8. ↑The Star. (2026). Election deepfakes now 'dangerously accessible' (complaint figures). https://www.thestar.com.my/news/nation/2026/11/13/election-deepfakes-now-dangerously-accessible-says-cyber-consumer-group
  9. ↑New Straits Times. (2025). MCMC mulls licensing rules for AI apps to curb deep fakes. https://www.nst.com.my/news/nation/2025/09/1272420/mcmc-mulls-licensing-rules-ai-apps-curb-deep-fakes