Deadline tracker · reviewed 2026-09-21
Every date that matters, with a source
20 obligations across the EU, Poland, South Korea and the United States. Status reflects the Digital Omnibus (Regulation (EU) 2026/1744) and the state-law changes of 2026. Machine-readable copy at /deadlines.json, and as JSON and CSV in the open-data repository on GitHub under CC BY 4.0; follow changes in any feed reader at /feed.xml.
Subscribe in your calendar: Apple / Outlook · Google Calendar · Outlook.com · .ics file. The calendar updates itself when a date changes.
In force applies now Upcoming a fixed future date Deferred postponed by legislation, still coming
By jurisdiction: EU · Poland · South Korea · United States. Terms explained in the glossary.
Each obligation also has its own page, for example Article 50 transparency. The data is free to reuse with attribution.
Upcoming and deferred
| Date | Obligation | Status | Who | What and evidence |
|---|---|---|---|---|
| 28 October 2026 in 23 days | KRiBSI inspections and fines begin Poland · Ustawa o systemach sztucznej inteligencji | Upcoming | Companies and public bodies operating in Poland | KRiBSI may inspect on 7 days' notice, remotely by default, covering documentation, procedures, internal logs, cloud systems and contractor agreements, and may impose the AI Act fine tiers. For SMEs the cap is the lower of the fixed amount and the percentage of turnover. Evidence: Inspection protocol, named contact, document index, trade-secret marking. Source: KTZR, Polish AI act summary |
| 2 December 2026 in 58 days | Machine-readable marking of generative AI output EU · EU AI Act as amended by Reg. 2026/1744 | Upcoming | Providers of generative AI systems and, in practice, companies publishing generated content | New obligation introduced by the Digital Omnibus: generated output must carry machine-readable marking. The same package bans generating non-consensual intimate imagery and child sexual abuse material, with penalties up to €35M or 7% of global turnover. Evidence: Marking mechanism in place for every generative system in the inventory. |
| 1 January 2027 in 88 days | Frontier-model developer duties US, New York · RAISE Act (as amended 27 March 2026) | Upcoming | Developers of frontier AI models only (trained with more than 10^26 operations); stricter duties for large frontier developers with annual revenue above $500 million | Transparency and reporting on training, deployment, safety protocols and incidents for frontier developers. The chapter amendment signed 27 March 2026 takes effect on 1 January 2027 and sets civil penalties of up to $1 million for a first violation and $3 million for subsequent violations. Not relevant to companies that only use AI. Evidence: Not applicable to deployers. |
| 1 January 2027 in 88 days | Colorado ADMT duties begin US, Colorado · SB 26-189 (Automated Decision-Making Technology) | Upcoming | Developers and deployers of automated decision-making technology affecting Colorado consumers | Replaces the 2024 Colorado AI Act. Requires pre-use consumer notices, an explanation within 30 days of an adverse outcome, meaningful human review rights and developer documentation. The earlier risk-management-programme and impact-assessment duties were removed. Evidence: Pre-use notice texts, adverse-outcome explanation procedure, human review process. |
| 1 January 2027 in 88 days | ADMT significant-decision obligations US, California · CPPA automated decision-making regulations | Upcoming | Businesses under the CCPA using automated decision-making for significant decisions about consumers | Notices, opt-out and access rights around automated decisions with significant effects. Evidence: ADMT notice, opt-out process, risk assessment where required. |
| 1 January 2027 in 88 days | Provenance duties extend to large online platforms US, California · AB 853 | Upcoming | Large online platforms (2027) and capture-device manufacturers (2028) | Extends SB 942 obligations to platforms from 1 January 2027 and to capture devices from 1 January 2028. Evidence: Platform provenance handling (platforms only). |
| 22 January 2027 in 109 days | End of the one-year grace period for fines (expected) South Korea · AI Basic Act | Upcoming | All businesses in scope of the Act | The grace period on administrative fines runs at least one year from 22 January 2026. Treat this date as the earliest point at which routine fines start. Evidence: All generative-AI notices and high-impact assessments complete. |
| 2 December 2027 in 423 days | High-risk obligations for stand-alone systems (Annex III) EU · EU AI Act as amended by Reg. 2026/1744 | Deferred | Providers and deployers of high-risk AI in employment, education, credit, essential services, law enforcement, migration and justice | Risk management, data governance, technical documentation, logging, human oversight, conformity assessment and registration. Deferred from 2 August 2026 by the Digital Omnibus. Evidence: Risk classification record, impact assessment, vendor documentation, oversight design. Source: Gibson Dunn, Omnibus agreement |
| 2 August 2028 in 667 days | High-risk obligations for AI in regulated products (Annex I) EU · EU AI Act as amended by Reg. 2026/1744 | Deferred | AI embedded in products under EU product law (machinery, medical devices, vehicles and similar) | Deferred from 2 August 2027 by the Digital Omnibus. Evidence: Product conformity files updated for AI components. Source: Gibson Dunn, Omnibus agreement |
Already in force
| Date | Obligation | Status | Who | What and evidence |
|---|---|---|---|---|
| 1 May 2024 887 days ago | AI disclosure duties in Utah US, Utah · Utah AI Policy Act (SB 149, narrowed 2025) | In force | Businesses using generative AI with Utah consumers, with stricter duties in regulated professions | Disclose the use of generative AI on request in consumer interactions and proactively in regulated professions. Evidence: Disclosure texts, script for staff answering 'am I talking to AI?'. |
| 2 February 2025 610 days ago | AI literacy duty, Article 4 EU · EU AI Act (Reg. 2024/1689) | In force | Providers and deployers of AI systems | Take measures to support the development of AI literacy of staff and others who operate or use AI on your behalf. The Digital Omnibus (July 2026) replaced the earlier duty to ensure, to your best extent, a sufficient level: no specific level has to be guaranteed and no certificate is required. The duty itself was kept, and national authorities supervise it. Evidence: An internal record of trainings and other measures (training register), the training programme and written AI usage rules; a short check of learning is good practice. |
| 2 February 2025 610 days ago | Prohibited AI practices, Article 5 EU · EU AI Act (Reg. 2024/1689) | In force | Providers and deployers of AI systems | Eight practices are banned: harmful manipulation and deception, exploiting vulnerabilities, social scoring, predicting crime from profiling alone, untargeted scraping of facial images, emotion recognition at work and in education (except for medical or safety reasons), biometric categorisation that infers sensitive characteristics, and real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions. Fines up to 35 million euro or 7% of worldwide annual turnover. The Commission's guidelines include recruitment in the workplace emotion ban. Evidence: Classification record showing that no system falls under Article 5; acceptable use policy that forbids the banned uses; vendors' written confirmation for tools with emotion or biometric features. Source: EU AI Act explorer, Article 5 |
| 2 August 2025 429 days ago | General-purpose AI model obligations EU · EU AI Act (Reg. 2024/1689) | In force | Providers of general-purpose AI models only | Technical documentation, copyright policy and a training-data summary for model providers, with extra duties for models with systemic risk. Companies that only use AI tools are not in scope. Evidence: Not applicable to deployers. |
| 1 January 2026 277 days ago | TRAIGA in force US, Texas · Texas Responsible AI Governance Act (HB 149) | In force | Developers and deployers of AI in Texas; state agencies | Bans intentional development or deployment of AI that incites self-harm or crime, AI-generated child sexual abuse material, non-consensual deepfake pornography and government social scoring. Rules for state-government AI. Evidence: Acceptable-use policy prohibiting the banned uses. |
| 1 January 2026 277 days ago | Generative AI training-data documentation US, California · AB 2013 (training data transparency) | In force | Developers of generative AI systems made available to Californians | Developers must publish documentation about the datasets used to train their generative AI. Evidence: Published training-data summary (developers only). |
| 22 January 2026 256 days ago | AI Basic Act in force South Korea · AI Basic Act | In force | Domestic and foreign AI developers and businesses using AI in products; foreign firms above revenue or user thresholds must appoint a Korean representative | Generative AI providers must notify users that AI is used and label AI-generated content, with clear notice where generated audio, image or video is hard to tell from real. One-year grace period before administrative fines, except serious cases. Evidence: User notices and labels on generative outputs, domestic representative if thresholds met. |
| 21 July 2026 76 days ago | High-impact AI obligations activated South Korea · AI Basic Act | In force | Operators of high-impact AI (healthcare, energy, transport, hiring, biometric analysis and similar) | Impact assessment on fundamental rights before deployment, meaningful explanation of outcomes, human oversight, a user protection plan and documentation. Fines up to KRW 30 million for notification failures, missing representative or refusing inspection. Evidence: Impact assessment record, oversight design, user protection plan. |
| 2 August 2026 64 days ago | Transparency obligations, Article 50 EU · EU AI Act (Reg. 2024/1689) | In force | Providers of AI that talks to people or generates content; deployers of emotion recognition, biometric categorisation, deepfakes and AI-written public-interest text | Chatbots must disclose they are AI unless obvious. Generated audio, image, video and text must be marked as artificial in a machine-readable way. Deepfakes and AI-generated public-interest text must be disclosed. Disclosure must be clear and given at the latest at first interaction. Not deferred by the Omnibus. Evidence: Disclosure texts in place, marking mechanism documented, transparency checklist completed per system. Source: EU AI Act explorer, Article 50 |
| 2 August 2026 64 days ago | Provenance and detection tools for large generative AI providers US, California · SB 942 (AI Transparency Act) | In force | Providers of generative image, video or audio tools with more than 1 million monthly users | Offer AI detection tools, watermark options and disclosures for AI-generated content. Delayed from 1 January 2026 to 2 August 2026. Evidence: Provenance features and disclosure options (large providers only). |
| 11 August 2026 55 days ago | Polish AI Systems Act, main provisions in force Poland · Ustawa o systemach sztucznej inteligencji | In force | Companies and public bodies operating in Poland | National law implementing the EU AI Act. Creates KRiBSI (Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji) as supervisor, a binding individual-opinion procedure (150 PLN fee, 30 to 60 days) and a regulatory sandbox that is free for SMEs. Evidence: Register of AI systems, assigned roles, AI usage policy, vendor commitments. Source: KTZR, Polish AI act summary |
Get an email before each date
One email 30 days and one 7 days before each date in the markets you tick, with who it applies to and what to have ready. Next: Machine-readable marking of generative AI output, 2 Dec 2026.
How to use this
- Run the readiness scan to see which rows apply to you.
- Copy the applicable rows into your AI inventory as the jurisdictions and obligations columns.
- Assign an owner and a review date to each row. The Compliance Kit includes the jurisdiction-mapping spreadsheet with all rows pre-filled.