1. Ministerial foreword 

Artificial Intelligence (AI) is one of the most vital technologies of our lifetimes. It has incredible potential to improve our public services, boost productivity and rebuild our economy. However, to take full advantage and fully realise these benefits we need to build trust in these systems which are increasingly part of our day to day lives. We must protect end-users and address the very real security threats to AI systems and models. Organisations in the UK must be confident they can adopt AI, and security must be built in across the AI lifecycle as a key safeguard against misuse. 

The voluntary Code of Practice on the cyber security of AI which is set out in this Government response will be used to inform the development of a global standard. The Code of Practice and the new implementation guide forms one-part of Government’s wider work on AI and is aligned and contributing to the vital programme that DSIT is progressing on frontier AI to prepare the UK for future advanced AI models. As announced in the King’s Speech this summer, we will deliver on our manifesto commitment by placing binding requirements on the handful of companies developing the most powerful AI systems. This highly targeted legislation will build on the voluntary commitments secured at the Bletchley and Seoul AI Safety Summits and strengthen the role of the AI Safety Institute. This work on the cyber security of AI is also aligned with DSIT’s other cyber security initiatives, such as the recently published draft Codes of Practice for Cyber Governance and Software Vendors which will both improve security practices, outcomes, and confidence for UK organisations. 

I greatly appreciate all the responses we received to the Call for Views on the cyber security of AI and the many contributions from international partners and industry. My officials have analysed your responses, and I am pleased to now introduce the government’s response to that Call for Views. 

This government response outlines how we have taken your feedback on board. I am delighted by the scale of support for DSIT’s approach and the technical feedback which has helped us to update the Code of Practice and create a brand-new implementation guide to support organisations in adopting it, particularly small and medium enterprises. We recognise it is vital that internationally agreed and aligned security requirements are developed and therefore my officials will be progressing with our plans to create a global standard.

We must ensure that all new and existing technologies are safely developed and deployed across the UK. The UK, as a world leader in securing technology, will continue to advocate the importance of cyber security and the need for a secure by design approach across all technologies. 

This is another step to ensure we can all benefit from secure AI, and I look forward to continuing discussions on how the government, international partners, industry and civil society can collaborate to achieve this goal. Thank you again for your contributions to this generation-defining technology.

Peter Kyle

Secretary of State for Science, Innovation and Technology.

2. Executive summary 

The UK is well positioned to take advantage of the range of benefits AI has to offer. However, through its recently published evidence base, the Government recognises there are clear risks to AI security which must be addressed so these benefits can be realised^{[footnote 1]}.

From the 15 May to 9 August 2024, DSIT held a Call for Views on the cyber security of artificial intelligence (AI). The Call for Views set out a proposed two-part intervention: the development of a voluntary Code of Practice, and to then use this as the basis for the development of a global standard focused on baseline cyber security requirements for AI models and systems. 

DSIT received 123 responses to the Call for Views. Most responses were from organisations, including industry associations on behalf of their members, as opposed to individuals. This was a global Call for Views, and we welcome the views received from a wide range of international partners. We are satisfied that we have gathered the views of a considerable number of relevant stakeholders. 

Most respondents to the Call for Views (80%) were supportive of DSIT’s proposed two-part intervention. There was also overwhelming support for the inclusion of each of the 12 individual principles contained with the Code (ranging from 83% to 90%). There were several recurring pieces of feedback received through the Call for Views. This included: the need for more detail or guidance on how to implement the Code; suggested changes to provisions within each principle of the Code; and suggestions for new provisions. Respondents also noted that the existing market might not provide the sufficient skills or capabilities need to implement the Code. The responses to open-text questions were rich in detail and varied widely, making it challenging to categorise the feedback into overarching themes. As a result, the themes identified are broad to ensure each theme captures feedback from multiple respondents. 

We have taken this feedback and used it to update the Code of Practice and create a new implementation guide that supports the Code of Practice. The guide provides detail that supports organisations, particularly small and medium enterprises (SMEs), with implementing the Code. We have updated the Code and created a new principle on end of life that covers the transferring of ownership of the training data and/or a model as well as the decommissioning of a model and/or system. We have also updated the stakeholder groups to include “data custodian” and “affected entities”, while addingmore clarity across the Code. The principles are now more contextualised to AI security risk, particularly where software requirements are referenced. We will be taking the updated Code of Practice and implementation guide into the European Telecommunications Standards Institute (ETSI) to develop a new global standard focused on baseline cyber security requirements for AI models and systems. We will continue to advocate an international approach, pursue our goal of increasing the adoption of security principles domestically and internationally and provide clarity to organisations on how they should protect AI technologies. 

This document provides a detailed overview of the feedback received from responses to the Call for Views. We have provided responses to each question to explain how feedback has been taken on board in updating the Code of Practice, and in determining the government’s next steps in this space.

3. Background 

AI continues to be one of this era’s most defining and powerful technologies and is increasingly part of our daily lives both at home and at work. Across a range of areas such as technology, finance, transport, agriculture, and crime prevention, AI is changing the way we work and interact with data. The UK AI sector itself is strong and growing, generating £14.2 billion pounds in revenue in 2023 alone. Moreover, it is highly productive, contributing an estimated £5.8bn in Gross Value Added (GVA) to the economy in 2023 and employing over 64,000 people in the UK.^{[footnote 2]} 

The UK government’s research^{[footnote 3]} into the cyber security of AI found that there are clear and specific risks to the security of AI models and systems throughout the AI lifecycle. It is therefore imperative that these are addressed so that millions of consumers and organisations can safely benefit from AI technologies. A Call for Views on the government’s proposed interventions was held from 15 May 2024 to 9 August 2024^{[footnote 4]}.

The proposals included a two-part approach, comprised of a voluntary Code of Practice for the UK, which forms the basis for the second part, a global standard developed at an international standards body. Together, these will establish baseline security requirements that will help reduce the number and impact of successful cyber attacks and therefore protect users’ data and the economy^{[footnote 5]}. 

This work is part of DSIT’s wider technology security programme and its secure by design approach across all digital technologies, which places the responsibility on those that develop technology to build robust cyber security into their systems. Due to overlap between the technology areas and stakeholders, this work on AI is closely linked to the government’s work on cyber governance and software security and resilience, as well as our other secure-by-design initiatives across consumer IoT, enterprise IoT, and App Stores. DSIT recently published a consultation response outlining approaches on the Cyber Governance Code of Practice. In a recent Call for Views, DSIT sought feedback on a proposed Code of Practice for software vendors, a response to this Call for Views is expected to be published soon. All cyber security Codes of Practice produced by DSIT are part of the government’s broader approach to improve baseline cyber security practices and increase cyber resilience in the economy. These Codes have been designed as part of a modular approach so that stakeholders can apply them in tandem depending on which technology areas are relevant to their business^{[footnote 6]}. 

This work also complements wider ongoing work across government to ensure the UK’s economy will fully realise the benefits of AI. This includes the government’s commitment to introduce highly targeted legislation for the handful of companies developing the most powerful AI models. These proposals will promote the safe development of AI, and support growth and innovation by reducing current regulatory uncertainty for AI developers^{[footnote 7]}.

4. Methodology 

The Call for Views was open to the public and stakeholders were able to respond via an online survey, written mail, or via email. All responses were analysed using the same methodology, following the removal of any duplicates. 

The Call for Views asked respondents 28 closed questions and 22 follow-up open text questions. The only mandatory questions in the survey required the respondent to provide demographic detail and to answer the initial question on whether they agree or not with DSIT’s proposed approach set out in the Call for Views. For some questions, respondents were offered the opportunity to expand on answers and provide more detail with qualitative open text boxes. These open text boxes were not mandatory. 

For open text response questions, all responses were reviewed and systematically analysed to identify common themes. Given the highly detailed and diverse nature of these responses, grouping them into overarching themes was often challenging. Consequently, the themes identified tend to be broad, and in many cases, are based on similar feedback from a relatively small number of respondents. When a particular theme emerged as the most frequently mentioned theme within a question, it has been highlighted in the summary below. If a theme was mentioned by 15% or more of respondents to a question, it has been categorised as a “frequently cited” theme. Themes mentioned by fewer than 15% of respondents have been classified as “less commonly cited” in the summary. 

The number of responses to the open text follow-up questions ranged from 27 to 65. An individual response to the open text questions could contain reference to more than one theme, where this has occurred all the themes from the response have been noted. Not all the open text responses to questions were relevant to the topic of the question. If they were relevant to other questions, then the response was considered and reflected in the analysis of that question. 

Please note that some of the percentages in this write-up do not sum to 100% due to rounding.

5. Overview of responses 

DSIT received 123 responses to the Call for Views. The majority (68%) of responses were from organisations, as opposed to individuals. The individuals who responded primarily identified themselves as cyber security/IT professionals (44% of individuals) or academics (20%). 

Many of the responses came from trade or membership bodies which were predominantly based in the UK and US, but with members from other countries. These incorporated the views of multiple stakeholders simultaneously. 

We are delighted that this was a global Call for Views, with responses received from the UK, the rest of Europe, the US, Japan, Singapore, the Republic of Korea, other parts of Asia and Oceania. Of the organisations that identified where they were from in their response, 54% were based in the UK, 26% in North America, 11% in Asia and 7% in Europe (excluding UK) ^{[footnote 8]}.

We are satisfied that we have gathered the views of a considerable number of relevant global stakeholders through this Call for Views and are grateful to the stakeholders that responded as well as those that helped promote the Call for Views^{[footnote 9]}. 

6. Section 1: Rationale & definitions 

6.1 Question 7 - In the Call for Views document, the government has set out our rationale for why we advocate for a two-part intervention involving the development of a voluntary Code of Practice as part of our efforts to create a global standard focused on baseline cyber security requirements for AI models and systems. The government intends to align the wording of the voluntary Code’s content with the future standard developed in the European Telecommunications Standards Institute (ETSI). Do you agree with this proposed approach? 

Response	Percentage of respondents
Yes	80%
No	11%
Don’t know	9%

The vast majority (80%) of the 116 responses to this question showed agreement with the proposed approach, with 11% opposing and 9% responding with ‘don’t know’. 

32 respondents provided additional evidence and reasons for their answer (both yes and no respondents), and some themes have been identified from these responses. However, these themes were only based on a small number of responses. A frequently cited theme from those that responded yes was the need to be aware of, and engage with, other international efforts (such as those by the US and EU). For those that responded no, the frequently sited themes were that a new standard is not needed and support for mandating security requirements immediately. Another frequently cited theme (from both yes and no respondents) was the need to provide more detail linked to the Code, such as sub-provisions and an implementation guide. 

Other, less commonly cited themes included: 

• Support for mandating requirements at some point in the future 
• Confusion over the modular approach put forward by DSIT and the number of cyber security Codes of Practice 
• The need for DSIT to conduct standards work in other standards development organisations.

6.3 Question 8 - In the proposed Code of Practice, we refer to and define four stakeholders that are primarily responsible for implementing the voluntary Code. These are Developers, System Operators, Data Controllers (and End-users). Do you agree with this approach? 

Response	Percentage of respondents
Yes	62%
No	29%
Don’t know	8%

The majority (62%) of the 109 responses to this question supported the approach, with 29% opposing and 8% responding with ‘don’t know’. 

65 respondents to this question who agreed with our approach regarding the four stakeholder groups provided further detail. The most frequently cited theme among respondents was the need for additional stakeholders to be added. This was also the most frequently cited theme among the 27 respondents who stated “no” to this question. 

Other, less commonly cited themes included: 

• The need to update/change the definition used for end-user 
• The need to update/change the definition used for data controller 
• The need to update the terminology of the 4 stakeholders 
• Disagreement (from respondents who answered ‘no’) with the 4 stakeholders as new terminology is needed. 

6.5 Question 9 - Do the actions for Developers, System Operators and Data Controllers within the Code of Practice provide stakeholders with enough detail to support an increase in the cyber security of AI models and systems? 

Response	Percentage of respondents
Yes	45%
No	32%
Don’t know	23%

45% of the 107 responses to this question supported the current level of detail provided in the Code. 32% believed the Code does not provide enough detail, and 23% of respondents responded with ‘don’t know’. 

48 respondents answered ‘yes’ to this question and provided further detail. The most frequently cited theme among these respondents was the need for additional detail including sub-provisions and guidance on how to implement the Code. This was also the most frequently cited theme among the 28 respondents who answered ‘no’ and provided further detail. 

7. Section 2: Code of Practice principles 

This section looked at gathering views on the 12 principles presented in the draft Code of Practice. Questions in this section presented each principle in full and asked whether the respondent supported the inclusion of the principle within the Code of Practice.10 

10 As part of this, respondents also had the opportunity to provide feedback on whether the requirements were shall, should or could/can. For clarification, standards development organisations define shall to mean that it is a requirement, should is a recommendation and could/can indicates where something is possible. Stakeholders that seek to adhere to the Code are expected to at least adhere to all the shall requirements. Support for the inclusion of each principle in the Code of Practice was very positive across each principle, ranging from 83-90% of respondents. 

Respondents were then asked follow-up open-text questions: 

Where respondents indicated ‘yes’, they were asked whether they had any suggestions on wording of any specific provisions in the principle in question. 

For respondents who answered ‘no’, they were asked to provide the reasons for their answer 

The additional feedback received via the open-text responses to both follow-up questions could be grouped into similar consistent themes across all 12 principles^{[footnote 11]}. 

There were multiple recommendations for changes to the wording of provisions and requests to remove certain provisions. This feedback has been captured under the theme ‘suggestions to changes to specific ‘provisions’. Also, there was feedback on the overall approach or focus of principles as a whole and these types of suggestions have been captured under the theme ‘suggestion of changes to the framing of the principle’. The full list of themes across the questions in this section were: 

• Suggestions of changes to specific provisions 
• Suggestions of changes to the title of the principle 
• The need for more guidance on implementing the principle/provision. 
• Suggestion of changes to the framing of the principle 
• Suggestion of new provision(s) 
• Principle is too burdensome/not practical. 

Secure design 

7.1 Question 10 – Principle 1: Raise staff awareness of threats and risks - Do you support the inclusion of Principle 1 within the Code of Practice? 

The vast majority (90%) of the 107 responses to this question supported the inclusion of Principle 1 in the Code, with only 5% opposed and 6% responding with ‘don’t know’. 

58 of the respondents who answered ‘yes’ and 6 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. The most frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. Another frequently cited theme was the suggestion of changes to provision 1.1. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 1.1.2, 1.2, 1.2.1, 1.3, 1.3.1 and 1.4 
• Suggestion of changes to the framing of the principle. 

7.3 Question 11 – Principle 2: Design your system for security as well as functionality and performance - Do you support the inclusion of Principle 2 within the Code of Practice? 

The vast majority (85%) of the 108 responses to this question supported the inclusion of Principle 2 in the Code, with only 9% opposed and 6% responding with ‘don’t know’. 

58 of the respondents who answered ‘yes’ and 8 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. The most frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. Other frequently cited themes included the suggestion of changes to provision 2.2 and 2.7. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 2.1, 2.1.1, 2.3, 2.4, 2.5 and 2.6 
• Suggestion of changes to the framing of the principle. 

7.5 Question 12 – Principle 3: Model the threats to your system - Do you support the inclusion of Principle 3 within the Code of Practice? 

The vast majority (84%) of the 106 responses to this question supported the inclusion of Principle 3 in the Code, with only 9% not supporting its inclusion and 7% responding with ‘don’t know’. 

57 of the respondents who answered ‘yes’ and 9 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. The most frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. 

Other, less commonly cited themes: 

• Suggestion of changes to specific provisions including 3.1, 3.1.1, 3.1.2, 3.2, 3.3, 3.4, and 3.5. 
• Suggestion of changes to the title of the principle 
• Suggestion of changes to the framing of the principle. 

7.7 Question 13 – Principle 4: Ensure decisions on user interactions are informed by AI-specific risks - Do you support the inclusion of Principle 4 within the Code of Practice? 

The vast majority (84%) of the 106 responses to this question supported the inclusion of Principle 4 in the Code, with only 8% not supporting its inclusion and 8% responding with ‘don’t know’. 

47 of the respondents who answered ‘yes’ and 8 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. The most frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. Other frequently cited themes included the suggestion of changes to provision 4.4 and 4.5, and changes to the framing of the principle. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 4.1, 4.2, 4.3 and 4.6 
• Suggestion of changes to the title of the principle 
• The need for more guidance on implementing the principle 
• Suggestion of changes to the framing of the principle 
• Principle is too burdensome. 

Secure development 

7.9 Question 14 – Principle 5: Identify, track and protect your assets - Do you support the inclusion of Principle 5 within the Code of Practice? 

The vast majority (88%) of the 104 responses to this question supported the inclusion of Principle 5 in the Code, with only 7% not supporting its inclusion and 6% responding with ‘don’t know’. 

50 of the respondents who answered ‘yes’ and 6 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. The most frequently cited theme among responses was the suggestion of changes to provision 5.1. Another frequently cited theme was the suggestion of the inclusion of new provisions within the principle. Finally, another frequently cited theme was the suggestion of changes to provision 5.3. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 5.2, 5.4 and 5.4.1. 
• Suggestion of changes to the framing of the principle. 

7.11 Question 15 – Principle 6: Secure your infrastructure - Do you support the inclusion of Principle 6 within the Code of Practice? 

The vast majority (87%) of the 102 responses to this question supported the inclusion of Principle 6 in the Code, with only 6% not supporting its inclusion and 7% responding with ‘don’t know’. 

49 of the respondents who answered ‘yes’ and 6 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. A frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. Another frequently cited theme was the suggestion of changes to provision 6.4. 23 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 6.1, 6.2, 6.2.1, 6.2.2 and 6.3. 

7.13 Question 16 – Principle 7: Secure your supply chain - Do you support the inclusion of Principle 7 within the Code of Practice? 

The vast majority (83%) of the 106 responses to this question supported the inclusion of Principle 7 in the Code, with only 8% not supporting its inclusion and 9% responding with ‘don’t know’. 

55 of the respondents who answered ‘yes’ and 7 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. A frequently cited theme among responses was the suggestion of the inclusion of new provisions within the principle. Other frequently cited themes included the suggestion of changes to provision 7.1, 7.2 and 7.2.1 and the need for more guidance on implementing the principle. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 7.3, 7.3.1 and 7.3.2. 
• Suggestion of changes to the framing of the principle 
• The principle is too burdensome. 

7.15 Question 17 – Principle 8: Document your data, models and prompts - Do you support the inclusion of Principle 8 within the Code of Practice? 

The vast majority (87%) of the 102 responses to this question supported the inclusion of Principle 8 in the Code, with only 6% not supporting its inclusion and 7% responding with ‘don’t know’. 

41 of the respondents who answered ‘yes’ and 5 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. A frequently cited theme among responses was the suggestion of the changes to provision 8.1.1. Another frequently cited theme was the suggestion of the inclusion of new provisions within the principle. Finally, another frequently cited theme was the suggestion of changes to the framing of the principle. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 8.1, 8.1.2, and 8.2. 
• Suggestion of changes to the title of the principle 
• The need for more guidance on implementing the principle 
• The principle is too burdensome. 

7.17 Question 18 – Principle 9: Conduct appropriate testing and evaluation - Do you support the inclusion of Principle 9 within the Code of Practice? 

The vast majority (88%) of the 104 responses to this question supported the inclusion of Principle 9 in the Code, with only 7% not supporting its inclusion and 6% responding with ‘don’t know’. 

50 of the respondents who answered ‘yes’ and 6 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. A frequently cited theme among responses was the suggestion of changes to the framing of the principle. Another frequently cited theme included the suggestion of changes to provision 9.2.2. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 9.2, 9.2.1, 9.2.3 and 9.3 
• The suggestion of new provisions within the principle 

Secure deployment 

7.19 Question 19 – Principle 10: Communication and processes associated with end-users - Do you support the inclusion of Principle 10 within the Code of Practice? 

The vast majority (87%) of the 101 responses to this question supported the inclusion of Principle 10 in the Code, with only 7% not supporting its inclusion and 6% responding with ‘don’t know’.26 

48 of the respondents who answered ‘yes’ and 5 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. Frequently cited themes among responses were the suggestion of new provisions within the principle, and the suggestion of changes to provision 10.1,10.2 and 10.3.2. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 10.3 and 10.3.1 
• The need for more guidance on implementing the principle 
• Suggestion of changes to the framing of the principle. 

Secure maintenance 

7.21 Question 20 – Principle 11: Maintain regular security updates for AI model and systems - Do you support the inclusion of Principle 11 within the Code of Practice? 

The vast majority (89%) of the 101 responses to this question supported the inclusion of Principle 11 in the Code, with only 5% not supporting its inclusion and 6% responding with ‘don’t know’. 

46 of the respondents who answered ‘yes’ and 4 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. Frequently cited themes among responses were the suggestion of changes to the framing of the principle and the suggestion of new provisions within the principle. 

Other, less commonly cited themes included:

• Suggestion of changes to specific provisions including 11.1, 11.2, 11.2.1 and 11.3. 

7.23 Question 21 – Principle 12: Monitor your system’s behaviour - Do you support the inclusion of Principle 12 within the Code of Practice? 

The vast majority (87%) of the 102 responses to this question supported the inclusion of Principle 12 in the Code, with only 7% not supporting its inclusion and 6% responding with ‘don’t know’. 

47 of the respondents who answered ‘yes’ and 6 of the respondents who answered ‘no’ provided an answer to the respective open text follow-up questions. A frequently cited theme among responses was the suggestion of changes to the framing of the principle. 

Other, less commonly cited themes included: 

• Suggestion of changes to specific provisions including 12.1, 12.2, 12.3 and 12.4 
• Suggestion of new provisions within the principle 
• The principle is too burdensome. 

8. Section 3: Further questions 

8.1 Question 22 - Are there any principles and/or provisions that are currently not in the proposed Code of Practice that should be included? 

Response	Percentage of respondents
Yes	50%
No	30%
Don’t know	20%

50% of the 100 responses to this question believed that there were other provisions that should be included within the code, with 30% believing there were not and 20% responding with ‘don’t know’. 

48 respondents who answered ‘yes’ to this question provided further detail. There were several less commonly cited themes, including: 

• The Code should include requirements focused on ethics 
• The Code should include requirements focused on data security 
• The Code should include requirements for areas or topics outside of security 
• The Code should provide more detail or guidance 
• The Code should include requirements on end of life/cessation. 

8.3 Question 23 - Where applicable, would there be any financial implications, as well as other impacts, for your organisation to implement the baseline requirements? 

Response	Percentage of respondents
Yes	62%
No	15%
Don’t know	23%

This question was only presented for respondents who identified themselves as responding on behalf of an organisation. The majority (62%) of the 65 respondents to this question responded with ‘yes’, with 15% responding with ‘no’ and 23% responding with ‘don’t know’. 

35 respondents who answered ‘yes’ to this question provided further detail. This follow-up question specifically asked respondents to provide data to support their response, however, very few responses provided this data. The most frequently cited theme among responses was the recognition that there would be costs or financial implications for their organisation to implement the baseline requirements. Another frequently cited theme was that implementing the baseline requirements would have capacity and capability implications for organisations implementing the Code. 

A less commonly cited theme included that there could be a skill gap in this space and that finding expertise will be challenging. 

8.5 Question 24 - Do you agree with DSIT’s analysis of alternative actions the government could take to address the cyber security of AI, which is set out in Annex E within the Call for Views document? 

Response	Percentage of respondents
Yes	64%
No	20%
Don’t know	15%

The majority (64%) of the 98 respondents to this question agreed with the list of alternative actions, with 20% opposing and 15% answering with ‘don’t know’. 

28 respondents who answered ‘no’ to this question provided further details. A frequently cited theme among these responses was support to mandate the security requirements. 

Other, less commonly cited themes included: 

• Support for guidance for industry stakeholders 
• Agreement with DSIT’s analysis that regulation would be burdensome, particularly for smaller companies.

8.7 Question 25 - Are there any other policy interventions not included in the list in Annex E of the Call for Views document that the government should take forward to address the cyber security risks to AI? 

Response	Percentage of respondents
Yes	49%
No	26%
Don’t know	26%

48% of the 101 respondents to this question believed there were other policy interventions not included in Annex E, with 27% answering with ‘no’ and 26% answering with ‘don’t know’. 

47 respondents who answered ‘yes’ to this question provided further detail. A frequently cited theme identified among responses was support for investment in developing skills, such as to implement AI cyber security. 

Other less commonly cited themes included: 

• The need to collaborate with stakeholders to ensure consistent development of policy 
• Support for the creation of an implementation guide 
• Support for a certification scheme to prove adherence to the Code 
• Support for UK government investment in more AI security research 
• Support for government to utilise procurement frameworks/processes to promote the Code.

8.9 Question 26 - Are there any other initiatives or forums, such as in the standards or multilateral landscape, that that the government should be engaging with as part of its programme of work on the cyber security of AI? 

Response	Percentage of respondents
Yes	54%
No	11%
Don’t know	34%

The majority (54%) of the 96 respondents to this question believed there were other initiatives or forums the government should be engaging with, with 11% answering ‘no’ and 34% answering ‘don’t know’. 

55 respondents who answered ‘yes’ provided further detail. The most frequently cited theme was the need to collaborate with other government entities. Another frequently cited theme was the need to engage with other standards development organisations. 

Other, less commonly cited groups that were suggested included: 

• Not for profit organisations 
• International organisations34 
• Membership/community interest bodies. 

8.11 Question 27 - Are there any additional cyber security risks to AI, such as those linked to Frontier AI, that you would like to raise separate from those in the Call for Views publication document and DSIT-commissioned risk assessment (which has published alongside the Call for Views document)? Risk is defined here as “The potential for harm or adverse consequences arising from cyber security threats and vulnerabilities associated with AI systems”. 

Response	Percentage of respondents
Yes	46%
No	53%
Don’t know	1%

The majority (53%) of the 99 respondents to this question answered with ‘no’, with 46% answering with ‘yes’ and 1% with ‘don’t’ know’. 

43 respondents who answered ‘yes’ provided further details. Responses to this question were very detailed and varied. Frequently cited themes to this question were risks linked to data and risks specific to AI cyber security risks, such as those associated with frontier AI. 

Other, less commonly cited themes included: 

• Risks linked to supply chains and employees 
• Crossover risks associated with different technologies.

8.13 Question 28 - Is there any other feedback that you wish to share? 

Response	Percentage of respondents
Yes	48%
No	52%

The majority (52%) of the 99 respondents to this question did not wish to provide any further feedback, with 48% responding with yes. 

48 respondents who answered ‘yes’ to this question provided further written feedback. Naturally responses to this question were varied, and many respondents thanked the government for its work in this area. One frequently cited theme was the need for the government to establish a mechanism for consulting industry on a continual basis. 

Other less commonly cited themes included: 

• Lack of clarity on how the Code aligns with other HMG publications 
• Offers to further collaborate with DSIT 
• More detail needed in the Code 
• The Code needs to be regularly updated/reviewed 
• Further changes suggested to the Code 
• Support for mandating security requirements 
• Requests that encouraged international standards and alignment.

E03283358 

978-1-5286-5409-8

---

This publication is licensed under the terms of the Open Government Licence v3.0 except where otherwise stated. To view this licence, visit nationalarchives.gov.uk/doc/open-government-licence/version/3. 

Where we have identified any third party copyright information you will need to obtain permission from the copyright holders concerned. 

This publication is available at www.gov.uk/official-documents. 

Any enquiries regarding this publication should be sent to us at AIcybersecurity@dsit.gov.uk 

ISBN 978-1-5286-5409-8 

E03283358 01-25