Our theory of change
Short version · Working draft, September 2026 · Farmed animals in the EU and UK, to 2030
A theory of change sets out what an organisation is trying to change, and the steps by which its work should get there.
Impact
AI does not entrench factory farming – it makes conditions for farmed animals visible and open to challenge.
Goal for 2030
AI used on farmed animals in the EU and UK is externally accountable: the models are tested for harm to animals, welfare data reaches regulators and the public, and non-compliance has consequences.
AIGAH works in all three layers of AI governance for animals – models, deployment and enforcement – and a fourth column, the hub, supports all three. In each column you can see what AIGAH does and produces, what that changes in what people know and think, and what changes in the end.
Model layer
What AI models must take into account · EU
What AIGAH does
Draft criteria and tests for the risks that general-purpose AI models pose to animals, with advisors and expert reviewers; brief the EU AI Office.
What AIGAH produces
A criteria document, reviewed by experts, with tests that AI developers could run on their models.
What people come to know and think
The EU AI Office and AI developers treat harm to animals as a risk that can be defined and tested.
What changes
The GPAI Code of Practice includes concrete criteria for harm to animals; AI developers test their models against them and report the results.
Deployment layer
What farms, slaughterhouses and fish farms must record and disclose · EU and UK
What AIGAH does
Document incidents; respond to consultations; draft criteria for certification and assurance schemes; prepare the case for the Commission's 2028 review of the AI Act's high-risk list.
What AIGAH produces
An evidence bank of harms to farmed animals from automated systems; consultation responses; criteria for certification and assurance schemes; a proposal for a welfare measure comparable across species; the evidence for adding animal uses of AI to the high-risk list.
What people come to know and think
Officials, advisory committees and certification and assurance schemes see AI used on animals as a documented risk, with criteria they can apply.
What changes
EU and UK law, and certification and assurance schemes, require operators using AI on animals to record welfare data against independently set indicators, and to disclose it.
Enforcement layer
AI evidence under existing law, starting with slaughterhouse CCTV · UK and Spain
What AIGAH does
Draft briefs for policymakers and industry on mandatory AI screening of slaughterhouse CCTV; approach the Food Standards Agency; find a partner organisation in Spain.
What AIGAH produces
Briefs on AI screening of slaughterhouse CCTV, and on how the legal duty of care applies when AI manages animals.
What people come to know and think
Inspectors and lawyers know that AI-flagged evidence exists, and that existing legal duties apply to animals managed by AI.
What changes
Enforcement bodies and lawyers use AI-analysed slaughterhouse CCTV to enforce welfare law that already exists.
The hub
Supports advocacy organisations in all three layers
What AIGAH does
Map where EU and UK rules leave animals out; draft templates; bring the field together.
What AIGAH produces
The regulatory gap map; templates; the newsletter; workshops.
What people come to know and think
Advocates know where the openings are and have text ready to send.
What changes
Advocacy organisations respond to AI-policy consultations in their own name.
Terms used
- GPAI Code of Practice
- The EU's code of practice for providers of general-purpose AI models under the AI Act. It names “non-human welfare” as an example of systemic risk but does not define it. Its first review is expected in 2027.
- General-purpose AI models
- AI models that can carry out a wide range of tasks, such as the large language models behind chatbots.
- EU AI Office
- The part of the European Commission that implements the AI Act and oversees the Code of Practice.
- The AI Act's high-risk list
- Annex III of the EU AI Act: the areas, such as critical infrastructure and education, in which AI systems carry extra obligations. It contains no animal context. By August 2028 the Commission must report on whether new areas are needed.
- Welfare measure comparable across species
- Sometimes called an interspecific welfare metric: a measure that allows comparison between species – for example, time spent in pain of different intensities – against which the welfare data operators disclose can be read.
- Certification and assurance schemes
- Private standards that farms sign up to, such as Red Tractor, RSPCA Assured and Soil Association in the UK, and the certification schemes used in fish farming.
- Duty of care
- Under the Animal Welfare Act 2006 (England and Wales), anyone responsible for an animal must take reasonable steps to meet its welfare needs.
- Slaughterhouse CCTV
- Mandatory in slaughterhouses in all four UK nations – England first, in 2018, Northern Ireland most recently, from August 2026 – and in Spain since 2023. In England the official vet must view at least 15 minutes of footage a day.
- Food Standards Agency
- The body whose official vets oversee animal welfare in slaughterhouses in England and Wales.
If you are interested in the extended version of our theory of change – with the full pathways, the assumptions behind each step and how we test them – write to us: hello@aigah.org.
← Back to AIGAH