UK's AI Training Copyright Fight: 11,500 Responses, Creative Industry Pushback, and a 2026 Deadline

Copyright and AI: what government teams need to prepare before March 2026

In December 2024, the Government opened a consultation on copyright and AI. It drew over 11,500 responses by February 2025 and triggered significant debate, especially over a proposal to let AI developers use copyright works for training unless rights holders opt out.

Parliamentary scrutiny during the Data (Use and Access) Act (DUAA) led to a compromise. By 18 March 2026, the Secretary of State for Science, Innovation and Technology must publish: (1) an economic impact assessment of the AI training options; and (2) a report on the use of copyright works in AI development. A progress statement has been published in the interim.

Where the consultation stands

The UKIPO reviewed just over 11,500 submissions (over 3,000 were template letters), with around 80 officials and analysts working through the responses. The headline numbers from the online survey are clear.

• 88% supported Option 1 - require licences in all cases.
• 7% supported Option 0 - make no changes to copyright law.
• 3% supported Option 3 (the Government's preferred option) - an exception for text and data mining with rights reservation (opt-out).
• 0.5% supported Option 2 - a blanket exception (no rights reservation).

Creative industry respondents heavily backed statutory transparency. Tech sector respondents were mixed on transparency and were more supportive of Options 3 and 2. The distribution reflects a large response from individual creators and creative sector organisations.

What must be in the March 2026 report

The final report must address all four AI training options and set out proposals on:

• Technical measures and standards to control use and access to copyright works in AI development.
• How copyright affects developer access to and use of data.
• Developer disclosure about use of and access to copyright works (e.g., training data summaries).
• Granting of licences to AI developers.
• Enforcement of rules on use and access to copyright works for AI.
• Treatment of AI systems developed outside the UK (notable in light of Getty Images v Stability AI).

Stakeholder engagement

Working groups are underway on:

• Control and technical standards: evaluating practical tools and standards (also being explored at EU level).
• Information and transparency: approaches like training data summaries.
• Licensing: strengths and gaps in current licensing frameworks.

Implications for public bodies

There's strong opposition to a broad text and data mining exception. Policy will need to balance clear protections for rights holders with support for AI-driven growth. Expect pressure for licensing-first approaches and tighter transparency.

• Procurement and contracts: Prepare for licence verification, evidence of lawful data sources, audit rights, and stronger indemnities in AI procurements.
• Transparency readiness: Anticipate requirements for training data summaries. Coordinate FOI responses and publishing strategies to avoid disclosing sensitive or third-party IP.
• Technical controls: Be ready to respect rights-reservation signals and metadata at scale (e.g., content credentials and machine-readable opt-out markers) across public datasets.
• Cross-border exposure: The report will address AI systems developed abroad. Build due diligence for overseas vendors and model providers into approvals and risk assessments.
• Research use: Clarify institutional positions on text and data mining in research settings and any reliance on existing exceptions.
• Stakeholder engagement: Map affected creators, collecting societies, vendors, and public sector researchers; feed practical issues and costs into the working groups.

Actions to take before March 2026

• Inventory and posture: Create a register of datasets and materials your organisation publishes that may be scraped or used for AI training; set a default stance (licence, opt-out, or open) and document it.
• Update templates: Add clauses on provenance, permitted uses, retraining/fine-tuning boundaries, model cards/training data summaries, audit, and offshoring restrictions to your AI procurement and grant templates.
• Pilot transparency: If you build or commission models, trial internal training data summaries and recordkeeping now. This will lower switch-over costs if disclosure becomes mandatory.
• Data hygiene and signals: Apply machine-readable rights notices and content credentials to high-value content and archives. Monitor how these are respected by vendors.
• Budget scenarios: Model potential licensing costs versus compliance savings across major AI projects and public data releases.
• Engage in standards work: Nominate technical and legal leads to the control/standards and transparency groups; bring practical case studies and edge cases.

Open questions to watch

• Whether any exception allows opt-out reservations or moves to licence-by-default.
• Standardisation of machine-readable opt-out and attribution (e.g., content credentials), and how this aligns with EU approaches.
• Where the line is drawn between pretraining, fine-tuning, and "use" for copyright purposes.
• Expectations on provenance, citation, and dataset documentation for high-capacity models.
• How enforcement will work for models trained outside the UK but deployed here.
• Interaction with data protection, database rights, and contractual access controls.

Helpful references

• UK Intellectual Property Office (consultations, guidance, and updates)
• EU AI Act (transparency and data governance touchpoints)

Build capability

If your team needs a fast primer on AI procurement, governance, and data usage, explore role-based training options here: Courses by job.