Present 

Members 

• Ms Helen Boaden (Chair)
• Mr Stephen Balchin
• Ms Vanessa Cuthill
• Mr Colin Godbold
• Professor Monica Magadi
• Ms Isabel Nisbet

Advisors

• Mr Rhys Nadin, UKSA

Secretariat

• Dr Pamela Calderon Ambrossen, UKSA
• Mr Nikhil Harsiani, UKSA
• Ms Stephanie Jacobs, UKSA

In attendance

• Mr Edward Bextor, UKSA (for item 3)
• Ms Rachael Colquitt, ONS (for item 4)
• Dr David Knott, Government Digital Service (for item 2)
• Mr Nick Mavron, ONS (for item 4)

Apologies

• Ms Nicola Shearman, UKSA

1. Minutes and matters arising from the previous meeting

1. Members of the Committee approved the minutes from the 23 January 2025 meeting.
2. The Chair acknowledged that this meeting will be the last for Mr Colin Godbold. She extended her personal and professional thanks to Mr Godbold, and noted that Mr Godbold has provided a keen attention to detail, interesting ethical challenge and a jovial spirit to the Committee. Mr Godbold will provide his reflections on his tenure as an NSDEC member in item 5 (NSDEC(25)24).
3. The Secretariat updated the Committee with progress on actions from the 23 January 2025 meeting. All actions were completed or in progress.

2. AI Ethics: Lessons and Case Studies from Industry and Government (NSDEC(25)21)

1. Dr David Knott, Government Chief Technology Officer, presented for this item.
2. The Committee members introduced themselves and provided their backgrounds. Dr Knott introduced himself, including his role at the Government Digital Service (GDS). He also briefly provided his professional background, which includes positions in commercial organisations.
3. Dr Knott said that he came to artificial intelligence (henceforth, ‘AI’) ethics having done his PhD in philosophy and having worked in technology companies. He outlined the three fundamental ethical approaches (deontology, utilitarianism and virtue ethics) and opined that many large technology organisations employ a utilitarian approach in their decision-making regarding emerging technologies.
4. Dr Knott provided a case study of a global bank in which he led on producing an AI governance structure in 2019. He explained that, while banks are highly regulated, there was little regulation around the use of AI. The principles he developed for the global bank included sensible use, fair treatment of customers and best practice of AI use in the field. The global bank had an existing clear chain of responsibility embedded into its governance structure, which could be used for establishing responsibility for AI use. The bank had an executive committee which Dr Knott said was open to receiving advice, which helped in embedding an AI governance framework in the company. Dr Knott also stated that his experience in the bank demonstrated that merely publishing a set of principles did not suffice; in addition, AI awareness training needed to be delivered to all staff. The outcomes of this training were that it allowed the bank to go faster with using AI, because they had the structures set up to allow people at all levels to legitimately speak up if they had any concerns.
5. Dr Knott spoke of a case study in the ‘big tech’ world. He explained that the tech company felt that they had to develop an AI governance framework because of increasing public interest in their use of AI methods. There was also less compliance of ‘big tech’ companies compared to financial sector organisations. Dr Knott outlined the ethical principles this tech company employed for AI, and they had the expertise to rigorously test for bias and harm. However, opposing views about the risks of AI use were not always well-received. Dr Knott stated that the company had a ‘staff philosopher’ who provided a deep-level knowledge base to the organisation on how to approach ethical issues.
6. Dr Knott informed that the GDS has established ethical principles for use of AI in government. While these principles are not yet consistent across government, they are available for departments to adopt. The GDS produced the Generative AI Framework for HMG in 2024, which includes ethical guidelines. The framework was updated at the beginning of 2025, which includes examples of non-generative AI use (referred to as ‘the playbook’). In discussion, the Committee raised the following:
   1. The playbook, as it stands, cites many ‘experts’ and industry professionals as contributors. The Committee would also like to see active public involvement in the curation of the playbook, so the public can play a significant role in the frameworks around AI. Dr Knott acknowledged that the playbook was, indeed, developed by consultation with academics and professionals only. He explained that in a policy context, public consultation would have been appropriate, yet in a digital context, it is in the first instance more apt to obtain the views of users of technologies, as opposed to the public’s. Dr Knott, however, recognised that GDS should do more public engagement to ascertain citizens’ pressing questions on AI.
   2. There needs to be an education campaign about the benefits and risks of AI, not just for students and professional users, but for the general public, too. Dr Knott thanked the Committee for their suggestion and explained that there is a relatively new team in the Department of Science, Innovation & Technology (DSIT) called ‘Digital Inclusion’. This team finds out the public’s access to networks, physical technologies and their digital literacy. Dr Knott said that he will follow up with DSIT’s Digital Inclusion team on their activities to increase AI literacy.
   3. Future iterations of the playbook should address how their framework for AI governance can be adopted by different industries (for example, academic, financial and public sector) given that each industry will have its own regulatory networks. Dr Knott acknowledged this point.
7. Dr Knott presented a summary of which directives he has found to be successful in AI ethics frameworks. These are: visible and sustained leadership; direct engagement and commitment of senior leaders; sticking to principles over time; genuine time spent on thinking about ethical issues; expert advice that preserves authority; and deploying an awareness/education programme for staff.
8. In further discussion, the Committee raised the following points:
   1. While risk is not the only consideration around which an ethical framework should be formed, risk is still a vital aspect. Dr Knott responded that he will consider GDS’ playbook including more about risks to, say, end users and society.
   2. The Committee are interested in particular work done on ethical principles of using AI in relation to vulnerable groups (especially children). Dr Knott agreed that vulnerable groups could be exploited by AI use, while recognising the utility of AI to build solutions that can, for instance, help children’s access to social care.
   3. Institutions should be clear about which types of users an ethical guidance piece is aimed at. Dr Knott agreed that more outward-facing guidance for end users would be helpful.
   4. There are examples from universities of good practice approaches to increasing ‘digital literacy’ amongst staff and students alike. Dr Knott said that the AI tools themselves can be very easy to use, but that institutions should be open that this is a surface-level type of ‘AI literacy’, and that there are more complex implications for users. He also said the government in their playbook might have to make clear the differences in utility between, for example, ChatGPT and search engines.
   5. Even within academic institutions, there are vast differences in how, and to what extent, AI use should be permitted in student and staff outputs.
9. The Committee stated that NSDEC will soon need to think about how the ONS acquires data which will be used in AI processes, and how to challenge assumptions that such data acquisition and use will always be in the public benefit.
10. The Committee thanked Dr Knott for his presentation and time. The Chair proposed that Dr Knott to return to present to NSDEC in 12 to 18 months, and Dr Knott welcomed this.

Action:

NSDEC Secretariat to arrange for Dr David Knott to present at an NSDEC meeting in 12 to 18 months’ time.

3. Data Ethics Self-Assessment Compliance Checks (NSDEC(25)22)

1. Mr Edward Bextor from the UKSA presented on this item. This item is a continuation of the UKSA’s ethics self-assessment compliance reviews, the most recent of which was presented to NSDEC in October 2024.
2. Mr Bextor gave an overview of the four projects he reviewed for compliance. He found three to be compliant, with one not progressing further after having had their ethics self-assessment signed off.
3. In October 2024, Mr Bextor recommended that projects update CADE regularly when their methodology changes (NSDEC(24)34, 5.3). He said that he still recommends this, as he has found a potential risk where projects make changes not in line with the original ethics sign-off, and they do not update CADE of these changes. He recommended that CADE formally establish a process where projects re-engage with the Data Ethics team upon meaningful changes to their methodology and/or research aims.
4. Mr Bextor explained that CADE could employ a similar process to that which UKSA’s Data Protection Team (DPT) use. The DPT have review cycles built in to their DPIA assessment system, which enable them to trigger conversations with projects should their processing of personal data change. This allows for the DPT to review such changes without the need to be ‘captured’ in a random sample of a compliance check, as is the current reliance for CADE.
5. In discussion, the Committee raised the following points:
   1. The language of ethics being seen as a ‘gateway’ suggests that an ethical review is a one-time thing, when it should be seen as an ongoing conversation. The Committee suggested using wording such as ‘illuminating’ instead. Mr Rhys Nadin agreed that the word ‘gateway’ might not be apt, and that Mr Bextor’s recommended process will enable CADE to give the necessary ‘illumination’ to projects at appropriate intervals.
   2. They acknowledged that ‘real-life’ limitations such as to resourcing mean that certain projects that have been given ethical sign-off can no longer go ahead. However, the decision to not having certain research done is in itself is a matter of ethical concern as well as the proposal to do it. By positioning ethical reviews as ‘continually illuminating’, there is a greater impetus for researchers to inform CADE of the ethical implications of their work not going forward.
   3. While CADE (and NSDEC) should have the ability to review the ethical implications of a project intermittently, researchers should also be able to challenge CADE’s findings, as well as the need to continually update CADE if they do not feel it proportionate to do so. Mr Nadin said he and Mr Bextor will take this point into consideration.
6. The Committee asked (in reference to one of the four compliance checks) if the ONS-based project took into account similar, non-ONS, surveys when describing the public benefit. Mr Bextor responded that he found in his compliance check that this project was aware of other European surveys and their publication schedules. This awareness affected their publication strategy such that they could maximise the public interest in their own survey’s release and, therefore, increase benefit to the public.
7. The Chair thanked Mr Bextor and Mr Nadin for their work.

Action:

CADE to explore adjusting the scope their current process to encourage and assist researchers to continually engage and update on CADE on changes, updates or stoppages.

4. Adopting an ‘indexing first’ approach to the delivery of data linkage services utilising the Reference Data Management Framework Indexes (NSDEC(25)23)

1. Ms Rachael Colquitt and Mr Nick Mavron, both from the ONS, presented on this item. This item has come to NSDEC before, most recently as an update item in the February 2024 meeting (NSDEC(24)06).
2. The research team presented an update on the ‘indexing first’ work. The ‘indexing first’ approach to the Reference Data Management Framework (RDMF) enables faster and more consistent data linkages. The research team also outlined the opportunities and challenges of this approach. They stated that they are working through questions of quality while they roll out the approach, initially in the Integrated Data Service (IDS).
3. The research team informed the Committee of what progress they had made since their presentation to NSDEC in February 2024. The ad-hoc research programme, which evaluates generalised methods of linkage against more bespoke methods, is ongoing. The work to address NSDEC’s request for more assurance on transparency, communications and engagement (NSDEC(24)06, 5.3.1) is progressing, and recommendations have been shared within ONS. The research methodology for the ‘indexing first’ approach has been given support by the ONS’ Methodology and Research Assurance Group, and the research team will next go to the Methodology Assurance Review Panel (MARP) (addressing NSDEC(24)06, 5.3.2). The ONS has also established a Data Linkage Quality Group to continually assess quality; the research team work closely with Administrative Data Research UK (ADRUK) to ensure they match users’ needs; and they work with the ONS’ internal linkage ethics team to ensure the safety and security of the data involved.
4. In discussion, the Committee raised the following points:
   1. They asked what happens if a data source, which is to be indexed, is subsequently discovered to have erroneous data. The research team responded that at this point, such a dataset could be redelivered or removed. They added that the quality metrics they continually obtain will show how precise the datasets are. These metrics will also help ascertain in which cases generalised indexing is not an appropriate method.
   2. It is unclear where the source of the deanonymisation ‘key’ will be in the ‘indexing first’ approach. The research team explained that a unique RDMF ID is attached to data when introduced to the index, and all personal data is removed. In addition, the research team have explored ‘tokenising’ this system such that the assigned unique ID would differ from project to project. This would reduce the risk of reidentification, which the Committee found to be a concern here.
   3. They observed from the paper that internal ONS experts and industry stakeholders have contributed to curation of the approach, yet there is little mention of engagement with the public or relevant think tanks. The research team responded that they engage with international partners, however, they recognised they could seek the views of wider research groups and/or representation from the public.
   4. The Committee recommended that the ONS produce educational pieces that can communicate to the public that their data is being indexed. The research team stated that they are producing explainer videos for this purpose, and that they will think about how to further communicate to the public.
   5. NSDEC requested clarification on why bespoke linkage would be more appropriate for small sub-groups than the generalised linkage. The research team explained that they have found that some populations are not well-represented in certain indexes, so this would result in datasets describing these populations having a low match rate if introduced into the generalised linkage. Therefore, in these cases, a bespoke linkage would yield a better match rate.
5. The Committee asked the research team about how long each dataset will be held for indexing. The research team clarified that the dataset will be held as an asset for as long as the memorandum of understanding (MOU) with the data owner states, with 1 year as being standard. They added that if RDMF find an asset to not be requested by users for some time, they will review the utility of its being kept, with an option to remove it from the ‘spine’.
6. Upon question from the Committee about which organisations can request the ONS to index, the research team stated that the ‘indexing first’ approach is currently only for internal ONS use and for asset introduction into the IDS.
7. The Committee recognised the use of NHS (healthcare) datasets in the indexing, and raised that, under the National Data Opt-Out (NDOO), citizens can request their patient records be excluded from use in research. The Committee asked what ONS’ practice will be if a person has ‘opted out’ and, if they do ‘honour’ an ‘opt-out’, how they will resolve the complication that people can modify their ‘opt-out’ position. The research team responded that they will need to speak to colleagues in the Data Acquisition team to clarify what the ONS’ practice will be.
8. The Committee thanked the research team for their work and time, and for their candour in their presentation. The Chair invited the research team to return with an update when appropriate, to which the research team accepted.

Action:

The RDMF team to provide NSDEC with details of how they will manage instances of healthcare dataset ‘opt-outs’ in their indexing.

Action:

NSDEC Secretariat to arrange for the ‘indexing first’ approach to return to NSDEC in a future meeting.

5. Reflections on NSDEC priorities (NSDEC(25)24)

1. The Chair welcomed all Committee members to provide their thoughts of NSDEC’s past operations. In particular, Mr Colin Godbold shared his reflections with the Committee given that this meeting will be his last as member.
2. Mr Godbold has been an NSDEC member for nearly 10 years, as has Ms Isabel Nisbet. Mr Goldbold remarked on how the technological landscape has changed during his tenure in the Committee. He commented on the effectiveness of the ethics self-assessment process, and the work of the NSDEC Secretariat. He opined that issues of public confidence in the government’s use of data will be of even greater import in the near future.
3. Mr Godbold recommended that new NSDEC members receive a comprehensive induction, particularly for lay members. He suggested that the Committee meet in-person at least once per year. He also impressed upon the Committee the need for meeting papers and applications to be in plain English, as he thought some documents brought to NSDEC still contain esoteric language. The Committee concurred that meeting papers should be shorter and written in simpler language.
4. Mr Goldbold gave his reflections on how the Committee might wish to ‘apply’ the UKSA’s ethical principles:
   1. On public engagement, he remarked that the ONS should improve their efforts in engaging with the public, and that they could use NHS’ recent efforts in public consultation as a good-practice example.
   2. There is often an assumption that most research will be in the ‘public benefit’, a judgment which is not always scrutinised. He recommended that the Committee should be more critical about whether a proposed project truly would benefit the public, or if it simply adds to ‘public discourse’.
   3. On the issue of individual agency, he said the Committee could more readily debate whether the purported value of proposed research reasonably accounts for the use of citizens’ data for which it was not originally collected.
5. Other Committee members contributed their reflections:
   1. Ms Nisbet commented that public opinion in the system of government checks and balances is currently changing.
   2. Ms Nisbet observed that the emergency context of the COVID-19 pandemic meant that some proposals which were deemed permissible at the time may now not be. She recommended NSDEC continue to think carefully about this distinction.
   3. On the nations of the UK, Ms Nisbet reflected that some research proposals which have some to NSDEC have not taken Welsh considerations into account, including Welsh language. She suggested the Committee ensure scrutiny of projects, which include use of Welsh data, that do not meaningfully take the concerns of Welsh people into account.
   4. Mr Stephen Balchin suggested that the Committee reflect on which instances NSDEC feedback has been most valuable.
   5. Members suggested that the UKSA ethical principles should be reviewed, with consultation from appropriate stakeholders.
6. The Chair thanked the Committee for their openness and, again, thanked Mr Colin Godbold for his valuable and influential service to the Committee.

Action:

NSDEC Secretariat to begin process of reviewing UKSA ethical principles.

6. Any other business

1. The Chair informed the Committee about the ongoing recruitment exercise for new NSDEC members, and urged current members share the vacancy advert with their networks.
2. Mr Nikhil Harsiani discussed the arrangements for the next meeting (in July 2025), in which the National Statistician, Professor Sir Ian Diamond, is due to attend.
3. Ms Stephanie Jacobs updated the Committee on some of CADE’s recent work, including contributing to the UNECE’s ethics reference book; continuing to streamline project application forms that apply using the Digital Economy Act 2017 (DEA); and working to redraft ONS policies such that ONS-based research will consult with the Data Ethics team at an earlier stage in their pipeline.
4. The Chair asked members if they permitted future NSDEC meetings to be recorded, which would aid the Secretariat in minute-drafting. All members approved of this.
5. The next meeting of NSDEC will be held on the week commencing 21 July 2025, date to be confirmed.

Action:

NSDEC Secretariat to ascertain members’ availabilities for the July 2025 meeting.