Artificial Intelligence Advisory Ethics Panel

POLICE AND CRIME COMMISSIONER FOR BEDFORDSHIRE

Artificial Intelligence Advisory Ethics Panel

Meeting of 29 April 2026

Held between 18:00 – 21:00

Minutes

Present:

Gareth John (GJ) – Chair, Independent Panel Member
Daniel Swain (DS) - Independent Panel Member
Tiggey May (TM) – Independent Panel Member
Vicky Withey (VW) - Independent Panel Member
John Tizzard (PCC) – Police and Crime Commissioner
Katie Beaumont (KB) - Head of Governance and Transparency - OPCC
Tarushi De Cruze (TDC) – Compliance Officer – OPCC
Daniel James (DJ) – Programme Manager - Bedfordshire Police
Rebecca Cumming (RC) – Business Change and Continuous Improvement Manager- Bedfordshire Police
Tristan Steyn (TS) – Business Change Manager - Bedfordshire Police
Ian Taylor (IT) – Chief Superintendent – Bedfordshire Police
Kim Coates (KC) – Sergeant – Bedfordshire Police

Welcome, Apologies, Quorum & Declarations

The Chair welcomed all attendees to the meeting, and all members of the panel introduced themselves.

Apologies were received from: Edward Braund, Jolene Neil and Neil Bradley.

The Chair assured the quorum was met.

DS declared a small personal shareholding in Palantir. He confirmed that the holding was minimal and did not confer any substantive influence.

The Panel noted the declaration. The Chair confirmed the interest was considered de minimis and unlikely to present a conflict but would be recorded given Palantir’s role as a service provider to the force.

PCC Introduction

The PCC thanked all for attending and the force. The PCC outlined his statutory role, confirming he has no operational responsibility for policing. His role is to set strategy and budget and to hold the Chief Constable to account, acting as a supportive but critical friend and representative of the public.

He highlighted his national responsibilities relating to data, technology and AI ethics and noted that the Panel’s work would inform both local and national governance. He thanked Panel members for their voluntary contribution and acknowledged support from PCC office staff and police colleagues.

The PCC recognised Bedfordshire Police’s national leadership in the ethical use of AI and stressed the importance of effectiveness, proportionality, public trust and adherence to the law. He noted that AI should support, not replace, human decision‑making.

He referred to governance arrangements for AI applications, including Live Facial Recognition, confirming his role is limited to strategic oversight and scrutiny rather than operational deployment.

The PCC requested permission to observe part of the meeting, reaffirmed the Panel’s independence, and expressed his support for its work.

Minutes, Actions & Matters Arising

Panel members approved the minutes of the previous meeting.

Action 1: TS provided an update.

Action 2: TS provided an update.

Action 3: Completed.

Action 4: Completed.

Action 5: Completed.

Under matters arising, the Chair referred to the IPCO inspection of the use of Nectar, which had been scheduled for the end of April, and asked whether there was an update.

DJ advised that the inspection took place the previous day in Hertfordshire Police, rather than Bedfordshire. However, the inspection involved the same application operating within the same environment as used by Bedfordshire. The outcomes of the inspection are expected to be available next week.

Action: To share the results of the Hertfordshire IPCO inspection to the Panel.

Approval of panel documents

Terms of Reference – Signed off and approved by panel members.

Evaluation criteria - approved by panel members with the expectation to add further criteria if needed.

Scoring Methodology –  DS noted a preference for avoiding a middle score to prevent scores clustering without clear judgement. TM advised that a hybrid scoring approach is often used, with both four‑ and five‑point scales depending on context.

The Chair emphasised that scoring must be evidence‑based, noting the absence at this stage of documentation such as DPIAs, legal approvals and bias testing. He stressed the need for clear audit trails, accountability and routes for redress. The Panel’s approach would be iterative, with expectations refined over time.

VW agreed, highlighting that AI use cases are unique and that continuous improvement should inform the Panel’s methodology.

The Panel agreed to adopt the proposed RAG thresholds and scoring framework as a starting point, subject to review and refinement.

Strategic Context

The Chair introduced strategic context as a standing agenda item, noting the importance of situating the Panel’s work within the wider legal, media and public scrutiny landscape, particularly in light of current developments affecting trust and confidence.

The Chair highlighted two key contextual issues:

• The Sean Thompson legal challenge relating to the use of LFR, noting that the court had found current use lawful at first instance, subject to appeal.
• Ongoing public and media scrutiny of Palantir, including debate around its wider associations and recent coverage regarding its use by the Metropolitan Police.

TM observed that concerns raised by the Police Federation appeared to relate more to how technology was being used rather than to Palantir itself as a supplier.

VW emphasised that public uncertainty is largely driven by a lack of explainability and transparency, particularly around purpose, deployment, data retention, data security, safeguards, and controls. She noted that while the public broadly recognises the benefits of AI, perceived gaps in visibility of governance, risk management and harm mitigation can undermine trust and increase legal and reputational risk if not clearly demonstrated.

The Chair reflected that it is important to distinguish between:

• the technology itself (including testing, evaluation and bias), and
• the force’s deployment, including internal processes, safeguards and public‑facing assurances.

He further highlighted the Panel’s role in scrutinising due diligence and procurement, including ensuring there is a clear audit trail demonstrating appropriate challenge to suppliers on issues such as testing, bias, accountability and safeguards.

DJ stressed the importance of panel independence and cautioned against being influenced by media narratives or issues arising from other organisations using the same suppliers. He noted that media reporting may lack full context and could skew perceptions if not treated carefully.

The Chair agreed, emphasising the need for the Panel to “stay in its lane” while remaining aware of wider contextual impacts.

DJ clarified for the record that Bedfordshire Police does not collaborate with the Metropolitan Police on technology use cases, including Palantir, and that systems, architectures, relationships and use cases are entirely separate.

AI Proposals  – Restricted

DJ provided an update of the system.

Challenges and questions from the panel:

The panel asked questions regarding the system. Bedfordshire Police provided responses to the questions. However, this information is not publicly accessible.

Evaluation of AI systems proposals

The Chair proposed that the Panel assess the submission by working through the evaluation criteria sequentially and agreeing collective scores, using the criteria document as a reference.

The Chair noted that key police AI governance evidence was not included within the submission pack, including confirmation that required governance steps (e.g. MPCC Responsible AI Checklist, AIPAS assessment, DPIA, FRIA and legal approval) had been completed. He asked whether these documents could be provided to support the Panel’s scrutiny, particularly in relation to legality.

KB advised that certain documentation could not be shared due to its official sensitivity and panel members not being vetted. She noted that the absence of data protection and information security colleagues had limited discussion on what could be shared and confirmed that further advice would be sought.

The Chair clarified that the Panel was not seeking legal opinions or to replicate internal approval processes, but required at a minimum written assurance that governance milestones had been completed and that legal justification was in place.

VW supported this approach, proposing the use of a checklist‑based assurance model, with written confirmation from appropriately independent and accountable stakeholders that required approvals had been obtained.

RC suggested that a less sensitive example or checklist could be shared to demonstrate how governance is applied in practice, subject to advice from Information Management and legal teams. She agreed to follow this up offline.

TM asked about the role of DPIA assurance. DJ confirmed that all DPIAs had been reviewed and signed off internally by Information Management, the Data Protection Officer and SIRO, and that the organisation had proactively engaged with the ICO in relation to Palantir, including prior to contractual arrangements.

The Chair concluded that, for future submissions, the Panel would require written confirmation that all required police AI governance steps had been completed and approved, including details of any conditions attached, to enable the Panel to fulfil its scrutiny role effectively.

Action: Written confirmation to be included in Panel submissions as to whether the proposed AI system has completed the police force’s relevant governance requirements and approvals (e.g. NPCC Responsible AI Checklist, AIPAS assessment, DPIA, FRIA and legal approval).

Legality & Compliance

Grading: 3

Comments:

The panel asked follow-up questions regarding legality and compliance.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Necessity & Proportionality

Grading: 4

Comments:

The panel asked follow-up questions regarding necessity and proportionality.

(Q) The chair asked whether non‑AI alternatives had been considered and documented.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Transparency & Accountability

Grading: 4

Comments:

The panel asked follow-up questions regarding transparency and accountability.

(Q) The Chair asked whether there are clearly defined policies, operational guidance and training materials governing the use of the AI system.

(Q) TM asked whether members of the public have access to a clear explanation of the AI system being used.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Fairness & Non-Discrimination

Grading: 5

Comments:

The panel asked follow-up questions regarding fairness and non-discrimination.

(Q) TM sked whether the AI system could be prompted to identify its own bias, similar to approaches used with general‑purpose AI tools.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Accuracy & Reliability

Grading: 4

Comments:

The panel asked follow-up questions regarding accuracy and reliability.

Action:  All submissions to provide written assurance confirming that appropriate supplier due diligence and testing had been conducted (e.g. security, compliance, accreditation).

Human Oversight

Grading: 5

Comments:

The panel asked follow-up questions regarding human oversight.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Privacy, Data Protection & Cybersecurity

Grading: 3

Comments:

The panel asked follow-up questions regarding privacy, data protection & cybersecurity.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Action: The information security officer to provide written assurance that all cyber security certifications have been completed.

Community Engagement

Grading: 4

Comments:

The panel asked follow-up questions regarding community engagement.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Action: To provide a written summary of engagement activity undertaken to date.

Ethical Review & Impact Assessment

Grading: 5

Comments:

The panel asked follow-up questions regarding ethical review & impact assessment.

(Q) VW sought reassurance regarding potential scope creep and whether data could be used for purposes beyond those intended.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Overall Score: 37 (Green)

DS observed that responsibility sits with the panel to define a clearer and more proportionate way of capturing evidence of compliance, noting that a wide range of documentation had been requested. He acknowledged that some raw documents may not be appropriate to share and that this concern was reasonable. He suggested that confirmation of completion, rather than full documentation, would often be sufficient, and that clearer mechanisms for requesting this information would avoid reliance on ad‑hoc discovery through discussion. He noted this as a shared learning point.

The Chair agreed and highlighted this as a key learning from the discussion. He stated that, to fulfil the panel’s scrutiny role, assurance should be evidence‑based where possible and permissible. At a minimum, the panel would expect written confirmation that key governance requirements, such as completion of the MPCC and AIPAS checklists and the securing of legal approval, had been met, without the need to review full underlying documentation.

Live Facial Recognition (LFR) – Force

IT provided a headline overview of the force’s use of Live Facial Recognition (LFR). Deployments commenced on 19 September 2025 and have since taken place on 39 occasions across the force area: 18 in Bedford, 18 in Luton Town Centre, two in Dunstable, and one in Leighton Buzzard.

All deployments operated at the nationally recommended algorithm threshold of 0.64. A total of 771,630 facial scans were undertaken, with biometric data being immediately deleted where no match was identified. This resulted in 121 true positive alerts and zero false alerts.

Of the true positives, 50 individuals were arrested (predominantly on warrant), eight were dealt with through alternative disposals, and the remainder related to intelligence checks where no enforcement action was required. A further 21 arrests arose through wider officer activity during deployments. Arrests primarily occurred in Luton and Bedfordshire, with none in Dunstable or Leighton Buzzard.

Watchlists ranged from 792 to 910 individuals. KC clarified that scan volumes reflect multiple scans of individuals moving through the recognition zone, rather than unique persons scanned.

It was confirmed that community engagement forms part of every deployment. No formal complaints or community concerns have been recorded to date, and informal queries have been addressed through on‑site engagement with officers.

Live Facial Recognition (LFR) – Panel

(Q) TM asked if any more thought had gone into the potential inclusion of victims or material witnesses on LFR watchlists, particularly where engagement could inadvertently place vulnerable individuals at risk, for example if a victim were accompanied by a perpetrator unknown to be aware of police involvement.

(A) IT confirmed that LFR has not been used for this purpose to date. He emphasised that any consideration of including a victim or witness on a watchlist would require exceptionally strong justification and Superintendent-level authorisation. He stated that all conventional lines of enquiry to engage the individual would need to be exhausted first, and that such use would be limited to high‑risk, exceptional cases. Any alert would be subject to dynamic risk assessment at the point of deployment, with careful consideration of the individual’s circumstances before engagement.

It was agreed that this approach mirrors stepped safeguarding protocols used in other sensitive areas of policing, with the clear objective of avoiding harm to victims or witnesses.

(Q) TM asked if a concise version of this rationale could be documented to support officers involved in LFR deployments.

(A) IT agreed, advising that this would be best reflected through policy caveats and bespoke operational briefings, with briefings recorded for audit and accountability purposes rather than reliance on general policy documents alone.

The chair invited the panel to consider whether sufficient information had been provided to enable a formal evaluation, noting gaps in supporting governance evidence, including legal advice, the DPIA, and completion of relevant AI governance checklists (AIPAS and MPCCC).

TM advised that further information was required, noting that it would be unfair to reach a judgement without it. She highlighted that explanations provided during discussion had materially changed her assessment and suggested that submitting targeted questions would enable the panel to complete its evaluation accurately.

VW agreed, stating that while she had strong contextual understanding of LFR safeguards, public engagement and legal background, there were too many gaps to complete the assessment confidently. She emphasised the importance of sufficient time and structured discussion to review each criterion and align scoring, and noted that documentation of assurance was key for the panel’s role.

DS suggested a proportionate approach to manage time, proposing that the panel collate a list of specific questions in advance, share these with the LFR team, and then undertake a focused discussion at the next meeting to support scoring.

Action: Panel members to review the Evaluation Criteria document and collate and submit a list of questions and evidence requirements to be used for the LFR submission and future submissions. Responses to be provided ahead of the next meeting.

The Chair proposed revising the existing submission form to align more closely with the panel’s nine evaluation criteria. He suggested restructuring the form to clearly prompt for key governance evidence, such as confirmation of legal approval, completion of the DPIA, and relevant AI governance checklists, to enable more efficient and focused evaluation. It was noted that this revised format could form a standard template for all future submissions.

VW supported the proposal and suggested that the revised template also allow space for contextual information, including relevant legal developments and wider external considerations that may inform the panel’s reasoning. It was acknowledged that such context is evolving and may require future adaptation of the criteria.

The Chair agreed that the immediate priority was to tighten alignment between submissions and evaluation requirements, while remaining flexible to future legal and policy developments. He noted that, based on discussion to date, the panel felt broadly assured, but that robust documentation was required to evidence this formally.

Action: To review and rework the existing submission form so that it is more tightly aligned to the panel’s nine evaluation criteria, including explicit prompts for required governance evidence (e.g. legal approval, DPIA, AIPAS/MPCCC completion).

Police / Panel Communications

KB informed the panel that the OPCC is drafting a press release following the first meeting to publicise the panel’s work and raise awareness of its role. She requested short quotes from panel members.

Horizon Scanning

The chair highlighted the importance of remaining sighted on emerging AI capabilities, the rapidly evolving policy and legal landscape, and relevant developments arising from recent legal cases.

The Chair also emphasised the value of learning from other forces and oversight bodies, recognising the panel’s role as an early pathfinder and the opportunity to promote consistency and share best practice over time.

KB confirmed that she is capturing the panel’s recommendations, scoring and actions for each submission, with a clear audit trail.

Action: KB to make a log of each of the evaluations.

TM offered to review ethics application frameworks used within her organisation to identify any relevant questions or considerations that may support the panel’s work, and to share these where appropriate.

AOB & Forward Plan

KB shared positive feedback received from the force, who commented on the value of the panel, the breadth of experience represented, and the constructive support provided. She noted that the panel was seen as a strong and valuable addition and thanked members for their time and commitment, particularly given the length of the meeting.

VW volunteered to review the existing submissions form with the evaluation criteria.

The Chair thanked all for attending.

POLICE AND CRIME COMMISSIONER FOR BEDFORDSHIRE

Artificial Intelligence Advisory Ethics Panel

Meeting of 29 July 2026

Held between 10:00 – 13:00

Minutes

Present:

Gareth John (GJ) – Chair, Independent Panel Member
Debra Allnock (DA) - Independent Panel Member
Tiggey May (TM) – Independent Panel Member
Vicky Withey (VW) - Independent Panel Member
Marcella Smith (MS) - Independent Panel Member
Katie Beaumont (KB) - Head of Governance and Transparency - OPCC
Tarushi De Cruze (TDC) – Compliance Officer – OPCC
Daniel James (DJ) – Programme Manager - Bedfordshire Police
Rebecca Cumming (RC) – Business Change and Continuous Improvement Manager - Bedfordshire Police
Jolene Neil (JN) – Information Governance Manager
Adam Crawley (AC) – Deputy Information Security Officer
Kim Coates (KC) – Sergeant – Bedfordshire Police
Matilda Person (MP) – Senior Governance and Compliance Officer - Cambridgeshire OPCC 

Welcome, Apologies, Quorum and Declarations

The Chair welcomed all attendees to the meeting, and panel members introduced themselves.

Apologies for absence were received from Edward Braund and Daniel Swain.

The Chair confirmed that the meeting was quorate.

No declarations of interest were made.

Minutes, Actions & Matters Arising

The minutes of the meeting held on 29 April 2026 were approved as a true and accurate record.

Action 1: Completed.

Action 2: Forms to be discussed under the relevant agenda item. Where supporting documentation cannot be provided, with a submission, an authorised officer will provide written confirmation on the Submission Form checklist that the relevant governance steps and documentation has been completed and approval granted, and / or attend the meeting to provide assurance.

Action 3: Forms to be discussed under the relevant agenda item.

Action 4: DJ provided an update.

Action 5: Completed.

Action 6: Completed. A basic submission relating to the underlying LFR system will be shared with panel members in advance of the next meeting on 28 October 2026.

Action 7: Completed.

Action 8: To be discussed under the relevant agenda item at this meeting.

(Q) DJ highlighted the importance of ensuring transparency within the Force and exploring how the panel's scoring and assessments could contribute to the National Register for AI Capabilities, enabling other forces to view local evaluations and risk assessments.

(A) The Chair acknowledged the potential link between the panel's work and wider national risk monitoring arrangements, whilst noting the importance of keeping processes proportionate and efficient. He suggested adding a column to the tracking log to record any conditions attached to approvals and comments, to improve the audit trail and ongoing monitoring of submissions.

Action: To add a ‘Conditions/Comments’ column to the tracking log.

Action: To prepare and send to the Panel for evaluation a submission relating to the underlying LFR system in advance of the next meeting on 28 October 2026.

Approval of panel documents

AI submission form with incorporated checklist

AC advised that he had only received the document shortly before the meeting and required additional time to review it fully.

VW explained that the revisions were intended to provide clearer guidance regarding the information required to support panel scrutiny.

TM stated that the revised format better reflected the AI ethical review cycle and removed unnecessary duplication, although further refinements may emerge once used operationally.

(Q) JN asked who would be responsible for completing the form and whether any approval process would apply.

(A) The Chair explained that the form would be completed by an authorised representative of the Force before being submitted to the panel. KB confirmed that submissions would pass through the relevant Force governance board before being presented to the panel.

(Q) TM suggested checking alignment against national AI policing principles including lawfulness, transparency, accountability, explainability and robustness.

(A) The Chair confirmed the form had been developed with reference to NPCC, AIPAS and related national guidance and such items were referred to in the checklist.

DJ advised that Bedfordshire, Surrey, South Wales and GMP were acting as national exemplar forces for AI governance. He noted that the panel's processes could inform future national standards and governance frameworks.

Action: Final comments on the submission form to be provided by mid-August 2026.

(Q) DA noted that the Live Facial Recognition submission form contained a number of empty fields and asked whether submissions with incomplete information should be returned for further completion.

(A) KB advised that it had previously been agreed that submissions would contain information that is readily available and proportionate to provide. As forms are completed by operational or tactical leads, some fields may occasionally be left blank. Where the panel requires additional information or clarification, this can be raised during the meeting as part of the scrutiny process.

(Q) AC asked whether the panel acted as an authorising body.

(A) KB clarified that the panel's role was advisory. Recommendations are provided to the Police and Crime Commissioner rather than being formal authorisation.

Strategic Context

The Chair highlighted continuing political focus on AI, including the introduction of a national Cabinet-level AI post under the new Prime Minister Andy Burnham.

VW referred to recent media coverage regarding AI sandbox environments and security breaches. She emphasised the importance of cybersecurity considerations when evaluating AI systems. The Chair noted that cybersecurity assurance had already been incorporated within the panel's revised submission form and checklist and members agreed that cyber security would remain a recurring area of scrutiny.

AI Proposals – Restricted

DJ provided an update of the system.

Evaluation of the system

Legality & Compliance

Grading: 4 (Subject to confirmation of legal approval and operational policy documentation.)

Comments:

The panel asked follow-up questions regarding legality and compliance.

(Q) The Chair referred to the Bridges judgment and asked whether operational guidance existed governing when the technology could and could not be used.

Action: To add to the submission checklist whether there has been legal approval.

Action: To provide panel members with evidence of legal approval for the tool, or confirmation from an authorised officer that such approval has been granted. In addition, to provide the operational manual, including the relevant guidance and controls referenced in the Bridges case.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Necessity & Proportionality

Grading: 4

Comments:

The panel asked follow-up questions regarding necessity and proportionality.

(Q) The Chair asked whether alternative options had been considered before implementing the AI-enhanced solution?

(Q) TM asked how data quality issues were identified and addressed.

(Q) DA asked whether specific testing had been undertaken regarding children and vulnerable individuals.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Transparency & Accountability

Grading: 4

Comments:

The panel asked follow-up questions regarding transparency and accountability.

(Q) The Chair asked how the Force communicates the use of the tool to the public and what arrangements are currently in place to ensure transparency and accountability.

(Q) VW asked whether there were plans to develop additional KPIs or performance measures that could help demonstrate the benefits of the tool, improve public confidence, and support positive communication around outcomes.

(Q) TM asked whether the technology had been independently assessed.

Action: To share the relevant independent assurance documentation with the panel.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Fairness & Non-Discrimination

Grading: 4 (Subject to summary information regarding testing for bias and discrimination).

Comments:

The panel asked follow-up questions regarding fairness and non-discrimination.

(Q) The Chair noted that, whilst the testing and retesting undertaken appeared robust, the panel's assessment must be evidence-based. He requested supporting information outlining the testing undertaken, the results achieved, identified risks, and the mitigation measures implemented, particularly in relation to potential bias and indirect discrimination.

Action: To provide a summary of the testing undertaken, including results and mitigation measures relating to bias and discrimination risks, where appropriate.

(Q) MS asked whether the technology could disproportionately affect people from ethnic minority backgrounds.

Bedfordshire Police provided responses to the above questions and other related matters. However, this information is not publicly accessible.

Accuracy & Reliability

Grading: 4 (Subject to assurance regarding supplier due diligence and procurement governance).

Comments:

The panel asked follow-up questions regarding accuracy and reliability.

Action: Provide written assurance regarding supplier due diligence and procurement governance.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Human Oversight

Grading: 5

Comments:

The panel asked follow-up questions regarding human oversight.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Privacy, Data Protection & Cybersecurity

Grading: 5

Comments:

The Chair requested assurance regarding DPIA approval and cyber security assessments. AC confirmed that cyber security reviews had been completed and approved. JN confirmed that DPIAs had been approved by the relevant governance authorities.

Members acknowledged that full DPIAs and security assessments may contain commercially sensitive or operationally sensitive information. Written confirmation from authorised officers may be sufficient assurance for panel purposes. The panel agreed that authorised written confirmation of approval in the Submission Form checklist would normally be acceptable where documents could not be shared.

Community Engagement

Grading: 4

Comments:

The panel asked follow-up questions regarding community engagement.

(Q) The Chair noted that the revised submission form includes a requirement for either a communications plan or a community engagement plan and asked whether any specific engagement activities had been undertaken in relation to this tool.

(Q) DA asked whether individuals would have the ability to challenge decisions made where AI had been used as part of the process.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Ethical Review & Impact Assessment

Grading: 4

Comments:

The panel asked follow-up questions regarding ethical review & impact assessment.

(Q) The Chair asked whether operational staff receive training not only on the use of AI enabled tools, but also on the ethical, human rights and public confidence considerations associated with their deployment. He noted the importance of ensuring staff are aware of potential public concerns surrounding the use of AI in policing.

Bedfordshire Police provided responses to the panel’s questions and other related matters. However, this information is not publicly accessible.

Overall Score: 38 placing it within the Green (Low Risk) category.

The panel supported continued deployment of the capability, subject to the provision of the assurances and supporting confirmations requested during scrutiny.

(Q) KB asked members to consider potential agenda items for the October meeting.

(A) Members agreed that the October meeting should include:

• A formal LFR submission of the underlying LFR system using the revised submission form and process.
• An update on Police AI and the National Register for AI Capabilities.

Live Facial Recognition (LFR) – Force

KC provided an overview of the LFR results since September 2025.

Live Facial Recognition (LFR) – Panel

(Q) MS sought clarification on the LFR performance data presented, specifically asking whether there had been any arrests or interventions resulting from false identifications.

(A) KC confirmed that there had been no false alerts, false identifications or resulting arrests since the first deployment of LFR in September 2025. She explained that deployment results are published on the Force website within seven days of each operation and include confirmation where no false alerts have occurred.

(Q) MS further asked whether all individuals stopped following an alert were subsequently arrested.

(A) KC explained that alerts can relate to a range of policing purposes, including arrest warrants, wanted persons, or intelligence-led interventions. Whilst some alerts result in immediate arrest, officers are still required to assess each case individually and satisfy themselves that the relevant legal thresholds are met before taking enforcement action.

(Q) MS raised the importance of maintaining public confidence, particularly within Black communities where concerns have previously been expressed regarding the accuracy of facial recognition technology. She asked whether ethnicity-related performance data was being collected and whether further public reassurance could be provided.

(A) KC advised that the Force has not yet published detailed ethnicity-specific performance data, although the underlying algorithm has been independently tested and deployment performance is continually monitored. She explained that there has been no evidence of disproportionate outcomes in deployment data to date. KC further advised that consideration is being given to undertaking further community engagement activity around the anniversary of the LFR deployment programme, including publishing a review of lessons learned and operational outcomes.

Recommendation: The panel recommended that, as part of the first anniversary of LFR deployments, the Force consider publishing a summary report on performance, outcomes, learning, and community engagement, including information that may assist in providing reassurance regarding equality, fairness and accuracy.

(Q) MS queried how images are obtained for individuals included on watchlists and whether photographs from non-police sources are used.

(A) KC confirmed that individuals would only be included where a lawful and proportionate image is available. She advised that police-generated custody images are generally used, and that the Force does not routinely seek photographs from third parties solely for LFR purposes. Individuals without a suitable image would not be included on the system.

(Q) The Chair asked whether the deployment of LFR at demonstrations or marches could have a "chilling effect" on individuals exercising their rights to freedom of expression and assembly. He queried the extent to which LFR is used at demonstrations and how concerns about data handling and public confidence are addressed.

(A) KC advised that Bedfordshire Police has not deployed LFR at demonstrations and would generally avoid doing so. She explained that proportionality and community impact are considered before deployment, and national APP guidance advises against use in such settings unless there is a clear policing justification. If an unplanned demonstration occurred, LFR would not be deployed in that location.

KC confirmed that LFR policies, Standard Operating Procedures, DPIAs and Equality Impact Assessments are published on the Force website and are subject to regular review.

Police / Panel Communications

(Q) The Chair asked whether the minutes of panel meetings would be published.

(A) KB explained that the minutes had not yet been published, as the panel first needed to review and agree them as an accurate record of proceedings. Going forward, approved minutes will be published on the PCC's website.

Horizon Scanning

The Chair referenced ongoing national discussions regarding AI governance, coordination between forces, and the potential development of a national AI register.

KB advised that Bedfordshire's AI governance arrangements and the work of the panel had been referenced during various national and Home Office meetings. She welcomed the planned update from DJ at a future meeting regarding developments in Police AI and the emerging national picture.

KB suggested that the Chair raise the panel's work during his forthcoming meeting with the PCC, noting that the PCC is the national lead for PCCs on AI and would be well placed to provide insight into emerging national developments and priorities.

The Chair agreed and invited panel members to provide any messages or issues they wished him to raise with the PCC.

AOB & Forward Plan

The Chair reflected on the importance of ensuring that the panel maintains a clear audit trail for the Force's use of LFR. Whilst the panel had received previous presentations and operational updates, he proposed that a formal submission on the underlying LFR system be brought to the 28 October 2026 meeting using the revised submission form. This should include the relevant supporting documentation or authorised confirmations relating to DPIAs, legal approvals and other governance requirements set out in the checklist on the form. Members agreed that this would strengthen the panel's scrutiny process and provide a robust record of assurance.

The Chair invited members to consider whether any additional specialist presentations would be beneficial to enhance the panel's knowledge and scrutiny capabilities.

KB thanked AC and JN for attending and contributing to the meeting at short notice. The Chair echoed these thanks and acknowledged the support provided by KB and TDC in facilitating the panel's work.

The Chair thanked all attendees for their contributions and closed the meeting.