Met Palantir pilot: The DPIA that raises more questions than answers

The high-risk processing the DPIA admits

The DPIA’s initial screening questions confirm the project was set to process special category data, criminal offence data, data concerning children under 13 and other vulnerable individuals, and data on a large scale. It includes tracking, monitoring, and surveillance in public areas, the use of AI, automated decision-making, and profiling of individuals. The assessment explicitly acknowledges the processing as “high-risk”.

In section Q2.25, the DPIA states that “the processing of personal data” would be considered “potential high-risk processing” in that it concerned “special category data, criminal offence data and vulnerable individuals … In particular, sensitive health data will be processed by collating and analysing the data for trends related to adverse sickness leave patterns.”

The scale is substantial. The assessment covers more than 50,000 current employees plus some former employees. It notes that “the processing involved in Foundry is heavily focused on the personal data, and actions, of MPS employees”, warning that if data were exfiltrated, it “could cause harm” and identify individuals.

Despite this risk profile, the DPIA acknowledges that multiple security and governance controls remained incomplete at the time of documentation. Automated access controls were marked “partially implemented”. Access logging was “planned but not yet in place”. Performance tests had not been planned. The system’s capability to manage data review, retention and disposal had “not been conducted”. No documentation existed for the flow and transformation of data throughout its lifecycle.

Computer Weekly received the DPIA on 3 June. The covering email described it as “an evolving document so continues to be updated”. The latest date entry in the DPIA is 31 March. 

The Senior Responsible Officer’s comments, dated 3 November 2025 – the date the DPIA was first written – acknowledge this tension directly: “While full testing has not been completed in the usual manner (i.e. deploying Palantir on a pre-production system first), I understand this risk has been partially mitigated through testing in situ. This reflects a balance between maintaining the speed of the pilot and ensuring appropriate safeguards. I am satisfied with the testing to date.”

The consultation gap

Perhaps the most striking single entry in the DPIA appears at question Q2.29, which asks whether the project team planned to consult with individuals whose personal data would be processed. The response was: “Considered and not required.”

This determination was made despite the project processing sensitive health data, sickness absence patterns, professional standards records from the Centurion system – a widely used platform for recording public complaints, conduct allegations, grievances and civil claims – and information relating to disabled employees, stop and search interactions, and custody-related issues.

The Met’s own Data Protection Officer (DPO) subsequently recommended precisely the opposite approach. In the DPIA’s Advice and Recommendations section, dated 31 March 2026, the reviewer states: “I recommend transparent communication with the individuals whose data will be processed. I recommend that a document explaining the purpose and scope of this project is produced and disseminated to MPS employees (staff and officers). I recommend producing an FAQ document with any concerns that may be raised by MPS employees (staff and officers).”

The DPO also highlights a purpose limitation concern: “The original purpose of collecting data for each system mentioned in the System Requirements spreadsheet is different to the envisaged purpose of this pilot. As such, this constitutes a change in purpose which will require telling MPS employees (staff and officers) of the change, as well as the lawful basis for processing.”

Computer Weekly put questions to the Metropolitan Police Federation regarding its consultation with the Met, the transparency of communications to staff, and concerns around data retention. No response was received.

What we do know is that Matt Cane, general secretary of the Metropolitan Police Federation, had told the BBC in April that staff “were never informed that the upgrade would include the deployment of Palantir’s artificial intelligence”.

Alternatives that were not evaluated

The DPIA states that alternatives to Palantir Foundry were “considered but not viable”, yet it names no other suppliers, provides no evaluation criteria, and offers no comparative assessment. The Met’s own in-house platform, the Enterprise Data Platform (EDP), is described in the DPIA as “an inhouse platform that ingests data”, with systems integration and data ingestion “ongoing under phase 2 of the Data, AI and analytics capability programme”.

The DPIA notes that Palantir offers “more enhanced capability than what EDP is currently able to do, and this pilot is validating that”. It also says Palantir was “faster and was able to produce an MVP to the timeframe that we needed. The AI capability offered by Palantir Foundry is not available with the EDP.”

The EDP was not yet capable of delivering the AI-driven analytics that Foundry could provide – but it was also conceived as the central architectural layer through which data would flow. The DPIA itself flags a “risk of lock-in with multiple data sources routing outside of EDP”, suggesting the Foundry pilot diverges from the Met’s own data infrastructure strategy.

A properly competitive process could potentially have evaluated other AI analytics providers in or alongside the EDP framework, but no such process is documented.

When Computer Weekly asked the Met to provide a list of other suppliers evaluated, the evaluation criteria used, and the business case detailing why Palantir was chosen, the force declined to engage further.

The references to a missing minimum viable product (MVP) are significant. As of mid-March 2026 – more than four months into a pilot originally scheduled to conclude in January – the core system had not reached a testable state. Throughout this period, data continued to flow into the platform.

The questions the Met did not answer

Computer Weekly submitted a second tranche of questions to the Met, addressing the gaps identified in the DPIA. These covered the absence of a competitive tender, the justification for sharing “not essential” data under the Bank Mellat proportionality test, the status of incomplete security controls, the evaluation of jurisdictional risk given Palantir’s status as a US-headquartered company, and the final disposition of MPS data held by Palantir after the pilot.

The Met’s response was a single sentence, declining to engage: “Following further conversations, we are not going to provide this level of detail and have nothing further to add to our previous statement.”

Several of these unanswered questions carry significant legal and operational weight. 

The DPIA’s own DPO section recommends that further questions be addressed, specifically: “What will Palantir do with the data after the trial ends (i.e., will Palantir retain a copy of the data or will any MPS data held by Palantir in relation to this trial be securely deleted)? If Palantir is planning to retain MPS data at the end of the trial, what is Palantir’s intended use of the data? If Palantir is planning to use MPS data to train its models, we assume that Palantir positions itself as an independent controller for this further processing.”

The DPO also notes: “It is not currently clear what will happen to the data at the end of the trial. If Palantir is planning to retain/use the data for its own purposes, it is important that we understand this so that we are able (a) to assess and articulate Palantir’s lawful basis for any further processing and (b) to be transparent with officers/staff about the fact that this will happen.”

On data minimisation, the DPO states: “As per the spreadsheet provided to do with System Requirements, there is a considerable amount of data to be shared with Palantir. This raises conflict with this principle, particularly around adequacy (the data collected should be sufficient to fulfil the intended purpose), relevance (only data that is directly related to the purpose should be processed) and necessity.”

The DPIA itself acknowledges at Q4.12 that data sharing with Palantir goes beyond the strictly necessary: “Due to the nature of the Palantir Foundry solution, and the broad nature of the use cases for the CSIE Palantir pilot, we are sharing additional data with the supplier in order to support the more holistic views/insights, etc for more proactive investigations.”

And on the fundamental question of purpose limitation, section Q5.06 is unambiguous. Asked whether personal data is being used strictly for the purposes for which it was originally collected, the response is simply: “No, the data is used for additional purposes not originally specified.”

A precedent for UK policing

The Met’s Palantir pilot does not exist in isolation. It sits within a broader landscape of UK police forces operating under the Police Digital Service and National Enabling Programmes model, predominantly on Microsoft-centric cloud infrastructure. The decisions made in this pilot – about procurement, transparency, data minimisation and jurisdictional exposure – are likely to reverberate across other forces considering similar deployments.

To what extent the DPIA’s internal recommendations have been resolved is unknown. The DPO calls for clear policy on data handling, training and awareness for staff using Foundry, auditing to enable full traceability and accountability, and clarity on what happens to data once the pilot ends. 

The Bank Mellat proportionality assessment notes that “if the objective is preventing misconduct, then analysing certain relevant data may be rationally connected”, but warns that “transferring all data we hold is not rationally connected unless we can show why each category of data is necessary”.

The Met’s refusal to engage with questions about alternative suppliers, jurisdictional risk and data disposal leaves significant governance questions unanswered. 

For a pilot that processes the personal data of more than 50,000 individuals – including health information, criminal offence data and information about vulnerable people – the gap between the speed of deployment and the maturity of safeguards is the story the DPIA tells, even when the Met will not.