MoJ: Check-In with your probation officer (E-Supervision)
AWS Rekognition is used within the Check-In with your probation officer service to support identity verification by performing facial comparison between user-submitted video and a photo held on file.
1. Summary
1 - Name
Check-In with your probation officer (E-Supervision)
2 - Description
The Check-In with your probation officer (E-Supervision) uses AWS Rekognition to help confirm a person’s identity when they complete a check-in.
As part of the process, a person on probation submits a short video. The system compares this to an existing reference image and provides a similarity score to support identity verification.
This tool is used to support practitioners in confirming that the person completing the check-in is the correct individual. It does not make decisions on its own, practitioners review the results and make the final decision
3 - Website URL
https://docs.aws.amazon.com/ai/responsible-ai/rekognition-face-matching/overview.html
4 - Contact email
checkinwithprobation@justice.gov.uk
Tier 2 - Owner and Responsibility
1.1 - Organisation or department
Ministry of Justice
1.2 - Team
Service Strategy and Transformation
1.3 - Senior responsible owner
Head of Transformation
1.4 - Third party involvement
yes
1.4.1 - Third party
Amazon Web Services (AWS) TPXimpact Ltd
1.4.2 - Companies House Number
TPXimpact Ltd – 06472420
1.4.3 - Third party role
Amazon Web Services (AWS): Provides the Rekognition service, supplying facial recognition capabilities used to generate similarity scores between images. TPXimpact Ltd: Provided delivery and implementation support for the Online Check-In service, including integration of AWS Rekognition into the service.
1.4.4 - Procurement procedure type
Framework agreement call-offs
1.4.5 - Third party data access terms
AWS processes image data submitted during check-ins for the purpose of generating similarity scores, in line with contractual terms and data protection requirements.
TPXimpact Ltd had access to systems and data as required to support delivery and implementation, under contractual agreements and in compliance with Ministry of Justice data governance and security policies. Neither TPXImpact Ltd nor AWS share data with any other parties or use it to train generative AI systems
Tier 2 - Description and Rationale
2.1 - Detailed description
The Check-In with your probation officer (E-Supervision) service uses AWS Rekognition to support identity verification for individuals on probation completing remote check-ins. The tool is integrated into a digital service that allows users to submit a short video as part of their check-in process.
From this submission, a still image is captured and compared against a reference image held on record. AWS Rekognition performs facial comparison and generates a similarity score indicating the likelihood that the images belong to the same individual. This output is used to support practitioners in assessing whether the correct person is completing the check-in.
The tool is intended for use by probation practitioners and service staff responsible for managing and monitoring compliance with supervision requirements. It is designed to operate as a decision-support tool and does not make autonomous decisions. Practitioners review the output alongside other available information and retain responsibility for the final determination.
The scope of the tool is limited to identity verification within the Check-In with your probation officer service. It is not used for broader surveillance, profiling, or automated enforcement decisions.
The performance of the tool may be affected by factors such as image quality, lighting conditions, and camera positioning. As with many facial recognition technologies, there may be variations in accuracy across different demographic groups. Where results are uncertain or confidence scores are low, additional checks or manual review are required.
2.2 - Benefits
The use of AWS Rekognition within the Check-In with your probation officer service supports the Ministry of Justice’s ambition to deliver more flexible, digital-first public services. It enables individuals on probation to complete check-ins remotely, providing greater accessibility and convenience while maintaining appropriate safeguards.
The tool supports practitioners by providing an additional source of information to help verify identity, contributing to more efficient and consistent decision-making. It also reduces reliance on in-person appointments, helping to optimise resource use and support service scalability.
2.3 - Previous process
Prior to the introduction of this tool, identity verification was conducted through in-person appointments or manual checks carried out by practitioners.
2.4 - Alternatives considered
A non-algorithmic alternative was considered to continue relying on in-person, face-to-face appointments for identity verification, without introducing the Check-In with your probation officer service. While this approach reduces reliance on automated tools, it is more resource-intensive, less flexible for service users, and limits the ability to deliver check-ins remotely at scale.
Algorithmic alternatives included other facial recognition and identity verification providers. These tools offer capabilities such as liveness detection and spoofing protection. AWS Rekognition was selected as it provides facial comparison capabilities, alongside features such as liveness detection and spoofing protection, and integrates effectively with existing infrastructure. It also allows similarity scoring to support human decision-making.
The chosen approach balances operational efficiency, scalability, and system integration, while maintaining practitioner oversight in final decisions.
Tier 2 - Deployment Context
3.1 - Integration into broader operational process
This tool forms part of the wider process for verifying that a person completing an Online Check-In is the correct individual. The system processes a video submitted by a Person on probation to perform facial comparison against a stored reference image. It then produces an outcome (e.g. pass/fail or flag), which is presented within the service.
Practitioners review all check-ins and use this output to support their assessment. Where a potential mismatch is identified, the system highlights this to the practitioner for further review. The tool contributes to, but does not determine, decisions about whether a check-in is valid.
3.2 - Human review
All check-ins are subject to practitioner review. Practitioners review submitted videos as part of their normal workflow, regardless of the system output.
Where the system identifies a potential mismatch, practitioners receive a notification and can carry out further checks or take appropriate follow-up action.
The tool acts as a decision-support mechanism by highlighting cases that may require additional scrutiny. Practitioners retain full responsibility for assessing identity and making any decisions.
Oversight is maintained through standard operational processes, guidance, and supervision within the service.
3.3 - Frequency and scale of usage
The tool is applied each time an eligible user completes an online check-in. Access to the service is subject to the approved eligibility criteria (currently people in risk Tiers D-G), meaning that not all individuals on probation use the Online Check-In service. As a result, the frequency and scale of usage are limited to the cohort of users who meet these criteria.
The scale of usage may change over time as eligibility criteria evolve and the service is expanded.
3.4 - Required training
Guidance on how to use the service is built into the user journey.
3.5 - Appeals and review
There is no formal appeals process, as the tool does not make decisions that impact access to the service. Where a check-in is identified as not matching, it is reviewed by a practitioner and does not prevent the individual from using the service. Any mistakes made by the recognition tool are not fed back to AWS.
Tier 2 - Tool Specification
4.1.1 - System architecture
The Check-In with your probation officer service integrates with AWS Rekognition via backend services to support identity verification.
A 5-second video is captured from the user during check-in, from which a still image is extracted. Both the video and the still image are stored in a UK-based Amazon S3 bucket managed by the Ministry of Justice.
The extracted still image, along with a reference image held on record, is sent to AWS Rekognition for facial comparison. Rekognition processes the images and returns a pass/fail outcome, which is used to support practitioner review.
4.1.2 - System-level input
The system receives a short video submitted by the user during the Online Check-In process and a reference image already held on record for the individual. It also uses session data generated by the backend service to initiate processing.
4.1.3 - System-level output
The system outputs one or more still images extracted from the submitted video, a confidence score generated during liveness processing, and a pass or fail outcome from the facial comparison against the reference image.
4.1.4 - Maintenance
The Ministry of Justice is responsible for maintaining the Check-In with your Probation officer service, including the backend and frontend integration with AWS Rekognition, storage of returned still images, and logging of operational issues.
AWS is responsible for maintaining the Rekognition service and its underlying models. The Ministry of Justice does not retrain or modify these models.
Monitoring is focused on issues encountered during use of the service rather than model retraining or direct maintenance of the AWS models.
4.1.5 - Models
The tool uses AWS Rekognition managed machine learning models for liveness detection and facial comparison. These models are provided and maintained by AWS. The Ministry of Justice does not train, fine-tune, or modify them.
Tier 2 - Model Specification
4.2.1. - Model name
AWS Rekognition
4.2.2 - Model version
n/a
4.2.3 - Model task
Facial comparison to assess whether two images are likely to contain the same individual
4.2.4 - Model input
Digital still images ( extracted frame from a user-submitted video)
4.2.5 - Model output
A comparison result indicating the likelihood of a match between two images
4.2.6 - Model architecture
N/A
4.2.7 - Model performance
N/A -
4.2.8 - Datasets and their purposes
N/A -
2.4.3. Development Data
4.3.1 - Development data description
N/A
4.3.2 - Data modality
N/A
4.3.3 - Data quantities
N/A
4.3.4 - Sensitive attributes
N/A
4.3.5 - Data completeness and representativeness
N/A
4.3.6 - Data cleaning
N/A
4.3.7 - Data collection
N/A
4.3.8 - Data access and storage
N/A
4.3.9 - Data sharing agreements
N/A
Tier 2 - Operational Data Specification
4.4.1 - Data sources
Operational data is sourced from user inputs. This includes video submissions provided by users during check-in, as well as reference images already held on record by the Ministry of Justice for identity match verification.
4.4.2 - Sensitive attributes
The operational data includes biometric personal data in the form of facial video and still images used for identity verification. While protected characteristics are not explicitly collected, they may be inferred from facial imagery.
Both video and still images are stored in a UK-based S3 bucket managed by the Ministry of Justice. Access is restricted to authorised personnel in line with data protection requirements.
4.4.3 - Data processing methods
N/A
4.4.4 - Data access and storage
User-submitted videos and extracted still images are stored in a UK-based Amazon S3 bucket managed by the Ministry of Justice.
Access to this data is restricted to authorised personnel, such as probation practitioners and service administrators, in line with role-based access controls and Ministry of Justice security policies.
AWS Rekognition processes still images for facial comparison and does not store the data.
4.4.5 - Data sharing agreements
Data sharing is limited to what is required for the operation of the service, and no data is shared beyond this purpose.
Tier 2 - Risks, Mitigations and Impact Assessments
5.1 - Impact assessments
Relevant impact assessments have been conducted as part of the development and deployment of the Check-In with your probation officer (E-Supervision) service. This includes a Data Protection Impact Assessment (DPIA) and Equality Impact assesment to assess risks associated with the processing of personal and biometric data.
These assessments identified key considerations relating to privacy, data security, and the potential for differential performance across demographic groups when using facial recognition technology.
Mitigations have been implemented through system design, practitioner oversight, and adherence to Ministry of Justice data protection and equality obligations.
5.2 - Risks and mitigations
Risk: Inaccurate or biased facial recognition results
Facial recognition technologies may perform differently across demographic groups or in suboptimal conditions (e.g. poor lighting or image quality), which could lead to incorrect matching outcomes.
Mitigation: The tool is used only as a decision-support mechanism. All outputs are reviewed by practitioners, and no decisions are made solely on the basis of the tool. Additional checks can be carried out where there is uncertainty. AWS continuously works to improve the performance and accuracy of Rekognition, including addressing known limitations in facial recognition technologies.
Risk: False positives or false negatives in identity verification
The system may incorrectly indicate a match or mismatch, potentially affecting how a check-in is assessed.
Mitigation: All check-ins are reviewed by practitioners, and system flags are used to support and not replace professional judgement.
Risk: Privacy and data protection concerns
The processing of biometric data, including facial video and still images, introduces privacy risks if not properly managed.
Mitigation: User-submitted videos are processed by AWS and deleted once processing is complete. The Ministry of Justice stores only the returned still image(s) required for service operation in a London-based S3 bucket, with access controls, secure storage, and contractual safeguards in place.