Quick answer
Since 2 August 2025, providers placing general-purpose AI models on the EU market, including Indian model builders, must keep technical documentation, give information to downstream providers, adopt a copyright policy and publish a summary of training content. Models presumed to have systemic risk (above 10^25 FLOPs of training compute) also need evaluations, incident reporting and cybersecurity. Non-EU providers need an EU authorised representative.
- GPAI model obligations have applied since 2 August 2025 to providers placing models on the EU market, wherever they are established.
- All GPAI providers must maintain technical documentation, inform downstream providers, adopt a copyright policy and publish a public training-content summary.
- Models trained with more than 10^25 FLOPs are presumed to have systemic risk and carry evaluation, incident reporting and cybersecurity duties.
- Non-EU GPAI providers, including Indian ones, must appoint an EU authorised representative.
- The GPAI Code of Practice, published July 2025, is a voluntary route to show compliance.
Who this is for: Indian foundation model and LLM builders, Teams releasing open-weight models, Model API providers, Startups fine-tuning and redistributing models, Research labs commercialising models
What counts as a general-purpose AI model?
The AI Act distinguishes AI systems (applications) from general-purpose AI (GPAI) models — models trained on large amounts of data, displaying significant generality and capable of performing a wide range of distinct tasks, that can be integrated into many downstream systems. Large language models, multimodal foundation models and large speech models typically qualify.
Why Indian model builders are in scope
Under Article 2, the Act applies to providers placing GPAI models on the EU market regardless of where they are established. "Placing on the market" covers making a model available through an API, a download, a cloud marketplace or integration into a product offered in the EU. An Indian lab releasing an Indic LLM that EU developers can download or call is likely a GPAI provider.
The two tiers
| Tier | Trigger | Obligations |
|---|---|---|
| All GPAI models | Meets the GPAI definition and placed on EU market | Technical documentation; information for downstream providers; copyright policy; public summary of training content |
| GPAI with systemic risk | Presumed above 10^25 FLOPs training compute, or designated | All of the above, plus model evaluations, serious incident reporting, cybersecurity protections and risk mitigation |
These obligations have applied since 2 August 2025. The Commission's AI Office supervises GPAI providers.
Baseline obligations for every GPAI provider
1. Technical documentation
Maintain documentation of the model — architecture, training process, data, compute, evaluation results — sufficient to provide to the AI Office and national authorities on request.
2. Information for downstream providers
Provide documentation to companies integrating your model into their AI systems, so they understand its capabilities and limitations and can meet their own obligations (for example, a downstream provider building a high-risk HR tool).
3. Copyright policy
Put in place a policy to comply with EU copyright law, including respecting rights reservations (opt-outs) expressed by rights holders for text and data mining. For crawled datasets, this means honouring machine-readable opt-outs and documenting how you do so.
4. Public summary of training content
Publish a sufficiently detailed summary of the content used to train the model, following the template provided by the AI Office.
Open-source exception
Models released under a free and open-source licence with publicly available weights and architecture benefit from exemptions from some of the documentation duties — but not the copyright policy or the training-content summary, and not if the model has systemic risk. Read the exact conditions before relying on it.
Practical tip
Build a model card pipeline that generates documentation, downstream information and the training summary from the same source of truth: dataset registry, training configs and evaluation results.
Systemic-risk models: the extra layer
A GPAI model is presumed to have systemic risk when the cumulative compute used to train it exceeds 10^25 floating-point operations. The Commission can also designate models based on other criteria. Providers must notify the Commission when a model meets the threshold.
Additional obligations
- Model evaluations, including adversarial testing, to identify and mitigate systemic risks.
- Assessment and mitigation of possible systemic risks at EU level.
- Serious incident tracking and reporting to the AI Office and, where appropriate, national authorities, together with corrective measures.
- Cybersecurity protection for the model and its physical infrastructure.
Will Indian models hit the threshold?
Most Indian models today — fine-tunes of open-weight bases, domain models, smaller Indic LLMs — are trained far below 10^25 FLOPs. Large sovereign foundation model efforts may approach it. Track cumulative compute, including continued pre-training, from day one so you know where you stand.
Fine-tuning and modification
If you fine-tune someone else's GPAI model, your obligations generally relate to your modification rather than the whole model, unless the modification is substantial. Commission guidance addresses where the line falls; check the AI Act page for current guidance.
The GPAI Code of Practice and the EU representative
Code of Practice
The GPAI Code of Practice, published in July 2025, sets out practical measures for transparency, copyright and, for systemic-risk models, safety and security. Signing and following it is voluntary, but it is the most straightforward way to demonstrate compliance to the AI Office. Non-signatories must show compliance by other adequate means.
For Indian builders, the Code's templates are useful even if you do not sign: they show the level of detail regulators expect in documentation and the training-content summary.
EU authorised representative
Providers established outside the EU must, before placing a GPAI model on the EU market, appoint an authorised representative established in the EU by written mandate. The representative:
- keeps a copy of technical documentation available to the AI Office;
- cooperates with authorities;
- can terminate the mandate if the provider acts contrary to its obligations.
The open-source exemption can relieve some providers of this duty for non-systemic-risk models; check conditions.
Choosing a representative
Pick an entity that understands AI documentation, not only a generic GDPR representative service. It will want to see your documentation before accepting the mandate, so prepare it first.
Training data: copyright, GDPR and DPDP
Model builders face three overlapping data questions.
Copyright (EU)
Your copyright policy must address text and data mining opt-outs. Document crawler behaviour, robots and metadata signals honoured, and how you handle takedown requests.
Personal data of EU individuals (GDPR)
Web-scale datasets inevitably include personal data. GDPR applies to processing of EU individuals' data, including training. Expect questions about legal basis, minimisation, filtering of special category data and handling of rights requests.
Personal data of Indian individuals (DPDP)
From 13 May 2027, DPDP obligations apply to digital personal data processed in India. Key points for model builders:
- Section 3(c) excludes personal data made publicly available by the Data Principal herself or by someone under a legal obligation. Not everything on the open web meets that test; scraped data republished by third parties may not.
- Data used under consent needs a notice naming training as a purpose (Rule 3).
- Erasure on withdrawal applies to datasets; see erasure requests in LLMs.
- Security safeguards under Rule 6 apply to training corpora containing personal data.
Read training LLMs on personal data under DPDP and DPDP and publicly available data before building a corpus.
Action plan for Indian model builders
- Decide whether you are a GPAI provider for each model and whether it is placed on the EU market (API, download, marketplace).
- Track training compute cumulatively against the 10^25 FLOPs presumption.
- Build documentation for authorities and downstream providers from a single source.
- Write a copyright policy with TDM opt-out handling and records.
- Publish the training-content summary using the AI Office template.
- Appoint an EU authorised representative unless an exemption clearly applies.
- Consider the Code of Practice and map your measures to it.
- Run privacy reviews for EU (GDPR) and Indian (DPDP) personal data in corpora.
- Monitor the Digital Omnibus proposal of November 2025 and Commission guidance for changes; treat proposals as proposals until adopted.
Common mistakes
- Assuming open-weight release removes all duties.
- Leaving training data provenance undocumented until a customer asks.
- Treating downstream documentation as marketing material.
- Ignoring the Act because the company is Indian.
For broader scoping, see EU AI Act for Indian companies.
Further reading on suhasbhairav.com
Before 13 May 2027
Compliance checklist
Classify each model
Decide whether it is a GPAI model and whether it is placed on the EU market.
AI Act Art. 2
Track cumulative training compute
Record FLOPs across pre-training and continued training against the 10^25 presumption.
AI Act (systemic-risk GPAI)
Maintain technical documentation
Document architecture, data, training, compute and evaluations for authorities.
AI Act (GPAI obligations)
Publish downstream information
Give integrators capabilities, limits and usage guidance.
AI Act (GPAI obligations)
Adopt a copyright policy
Honour TDM opt-outs and keep records of crawler behaviour.
AI Act (GPAI obligations)
Publish the training summary
Use the AI Office template for the public training-content summary.
AI Act (GPAI obligations)
Appoint an EU authorised representative
Give a written mandate to an EU-established representative before release.
AI Act (GPAI obligations)
Review the Code of Practice
Map your measures to the July 2025 GPAI Code of Practice.
GPAI Code of Practice
Run DPDP review on Indian data
Check s.3(c) applicability, notices and erasure handling for Indian personal data.
DPDP s.3(c); Rules 3, 6
Key dates
EU AI Act enters into force
GPAI model obligations apply
DPDP core obligations apply to Indian personal data in training corpora
Answers
Frequently asked questions
Do EU AI Act GPAI rules apply to Indian LLM builders?
Do EU AI Act GPAI rules apply to Indian LLM builders?
Yes, if they place a general-purpose AI model on the EU market, for example via an API, download or marketplace accessible in the EU. Article 2 applies regardless of where the provider is established. Obligations have applied since 2 August 2025.
What must every GPAI provider do?
What must every GPAI provider do?
Maintain technical documentation, provide information to downstream providers integrating the model, adopt a policy to comply with EU copyright law including text and data mining opt-outs, and publish a sufficiently detailed public summary of training content using the AI Office template.
What is a systemic-risk GPAI model?
What is a systemic-risk GPAI model?
A GPAI model presumed to have high-impact capabilities, primarily when its cumulative training compute exceeds 10^25 floating-point operations, or one designated by the Commission. Providers must add model evaluations, systemic risk mitigation, serious incident reporting and cybersecurity protections.
Are open-source models exempt from GPAI obligations?
Are open-source models exempt from GPAI obligations?
Partly. Models released under a free and open-source licence with public weights are exempt from some documentation duties and the authorised representative requirement, but must still have a copyright policy and publish a training-content summary. The exemption does not apply to systemic-risk models.
What is the GPAI Code of Practice?
What is the GPAI Code of Practice?
A voluntary code published in July 2025 setting out practical measures for transparency, copyright and safety and security. Following it is a recognised way to demonstrate compliance with GPAI obligations; providers that do not sign must show compliance by other means.
Do Indian model builders need an EU representative?
Do Indian model builders need an EU representative?
Yes, non-EU providers must appoint an EU-established authorised representative by written mandate before placing a GPAI model on the EU market, unless an exemption such as the open-source one applies. The representative holds documentation and cooperates with the AI Office.
Legal references and sources
- EU AI Act, Regulation (EU) 2024/1689 — GPAI model obligations (Chapter V of the Act) (official text)
- EU AI Act, Regulation (EU) 2024/1689 — Article 2 (scope) (official text)
- Digital Personal Data Protection Act, 2023 — Section 3(c) (publicly available data) (official text)
This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.
About the author

Suhas Bhairav
Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.
Related guides
- EU & cross-borderEU AI Act for Indian Companies: Does It Apply to You?A scoping guide for Indian companies: when Regulation (EU) 2024/1689 reaches you, which role you play, which obligations and dates apply, and what the Digital Omnibus proposal might change.
- AI & personal dataTraining LLMs on Personal Data Under the DPDP ActTraining or fine-tuning a model on personal data is processing under the DPDP Act. Here is how lawful basis, public data, research, erasure and memorisation work.
- DPDP Act 2023DPDP and Publicly Available Data: What Scrapers and AI Teams Must KnowSection 3(c)(ii) excludes personal data made public by the Data Principal herself or under a legal obligation. What that means for web scraping, AI training and data enrichment.
- EU & cross-borderAI Act Article 50 Checklist for Indian GenAI StartupsA task-level checklist for Indian GenAI startups to meet EU AI Act Article 50 transparency duties: chatbot disclosure, machine-readable marking, deepfake and public-interest text disclosure.
- EU & cross-borderSelling AI SaaS to Europe From India: Complete Compliance GuideThe full compliance stack for an Indian AI SaaS company selling into the EU: GDPR scope and roles, DPAs and SCCs, EU AI Act roles and dates, and how it fits with DPDP at home.
- EU & cross-borderDoes India Have GDPR Adequacy? What It Means for Data TransfersIndia has no EU adequacy decision. What that means for transfers of EU personal data to India, the tools that fill the gap, whether DPDP changes the picture, and how to design around it.