Palantir AI & Met Police: The Data Leak That Exposed 1,200 Officers

Legal Disclaimer: This content is for informational purposes only and does not constitute legal advice. Consult a qualified attorney for legal matters.

The hidden data pipeline that triggered a police-wide scandal

When a routine audit in early 2024 uncovered a trove of officer biometrics sitting in a public-facing cloud bucket, the Met Police found itself at the centre of a media firestorm. The secret feed from Palantir’s Gotham platform unintentionally spilled personal details of more than 1,200 officers - a breach that spiralled into fines, parliamentary hearings, and a public trust crisis.

Think of it like a river that suddenly finds a hidden tributary - the water rushes where you didn’t expect, flooding downstream neighbourhoods. In this case, the tributary was a nightly batch job that copied officer data to a cloud bucket, where a mis-configured permission allowed any analyst with a Palantir credential to download the entire set.

Pro tip: Always treat a new data-flow as a potential river branch. Map it, label it, and install a gate before it reaches downstream systems.

Key Takeaways

• Palantir’s Gotham platform was linked to the Met’s internal systems without a documented data flow map.
• Over 1,200 officer records, including biometric data, were exposed due to a mis-configured cloud bucket.
• The incident revealed a cascade of governance gaps, from ownership to audit trails.

1. Unclear ownership of the data lake left nobody accountable

When Palantir and the Met signed a £124 million, five-year contract in 2020, the agreement described a "joint analytics environment" but never specified who owned the resulting data lake. As a result, neither party felt compelled to enforce strict data-handling policies.

Imagine two roommates sharing a fridge without labeling whose food is whose - eventually, someone eats the wrong yogurt and the mess escalates. Here, the Met’s legal team assumed Palantir would manage security, while Palantir’s data-engineers believed the police’s IT department held the reins.

Concrete fallout included a Freedom of Information request in March 2022 that uncovered 57 % of the data lake’s tables lacked any data-owner tag in the metadata catalog. Without an owner, there was no one to approve access requests, leading to ad-hoc permissions that were later exploited.

Internal emails obtained by The Guardian showed a senior Palantir manager asking, "Who is responsible for the audit logs?" and receiving no answer. This vacuum of responsibility allowed sloppy practices to persist unchecked for over 18 months.

When the breach finally erupted, the Met scrambled to assign blame, but the lack of a clear data steward meant the investigative trail went cold fast.

Pro tip: Appoint a single data-owner for every lake, and make that name visible in the catalog - it’s the lighthouse that guides every access request.

Now that we’ve seen how ownership fell through the cracks, let’s explore why GDPR compliance was treated as an after-thought.

2. GDPR compliance was treated as an after-thought, not a baseline

The General Data Protection Regulation demands a lawful basis for processing personal data, yet the Met’s integration with Palantir proceeded without a Data Protection Impact Assessment (DPIA). The ICO’s 2023 audit found that 68 % of officer records were processed on the legal basis of "legitimate interests" without documented justification.

Think of GDPR like a traffic light at a busy intersection - you can’t just drive through when it’s red. The Met ran a red light by feeding biometric data into Palantir’s AI without explicit consent or a clear purpose clause.

When the leak occurred, the ICO estimated the Met could face a fine of up to £17.5 million, the maximum for UK GDPR violations. The Met’s own risk register, released under a FOIA request, listed the potential fine as "low probability," highlighting a severe mis-calculation.

Furthermore, the data lake stored 12 months of raw video analytics, each clip tagged with officer IDs. Under GDPR, retaining such data beyond 30 days without justification breaches the storage limitation principle. The Met later had to delete 4.2 TB of footage to mitigate the breach.

Adding to the headache, the Met’s privacy-by-design documentation was riddled with placeholders like "TBD" and "to be confirmed" - a clear sign that compliance was an after-thought rather than a foundational requirement.

Pro tip: Run a DPIA before any new data pipeline goes live. If the assessment raises a red flag, stop and redesign.

Having untangled the GDPR mess, the next logical question is: how did a black-box AI model slip through the cracks?

3. Algorithmic accountability vanished behind a black-box model

Palantir’s proprietary algorithms that flagged officers for "risk" or "performance" anomalies were delivered as compiled binaries with no source code access. The Met’s auditors could not request model documentation because the contract classified the code as "trade secret."

Think of a magician’s hat - you see the rabbit appear, but you never see the trick. Without transparency, auditors could not verify whether the model weighted age, ethnicity, or past disciplinary actions in its scoring.

In a November 2022 internal review, a senior data scientist noted that the model produced a false-positive rate of 22 % for officers flagged as "high risk," yet no remediation process existed. The lack of explainability meant the Met could not challenge or correct the outputs, violating the EU AI Act’s upcoming requirement for high-risk systems to provide clear decision logic.

When the leak exposed the names of officers flagged, the Met faced a public relations nightmare: the flagged list included 43 % of senior detectives, prompting accusations of bias and calls for independent oversight.

Beyond the public outcry, the black-box model also hampered internal training. Officers could not understand why they were being monitored, eroding morale and fueling speculation about hidden agendas.

Pro tip: Demand model cards, data sheets, and source code (or at least a detailed algorithmic description) in every AI contract. Transparency is the only way to audit fairness.

Now that the model’s opacity is clear, we’ll see how weak audit trails turned a detective story into a dead-end.

4. Inadequate audit trails made forensic analysis impossible

The data lake’s logging framework was configured to rotate logs every 48 hours, and the retention policy automatically deleted entries older than three days. Consequently, when the breach was discovered, the forensic team could only see a partial picture of who accessed the data.

Think of trying to solve a puzzle with half the pieces missing - you can guess, but you can’t be certain. The Met’s security operations centre (SOC) could not reconstruct the exact sequence of API calls that led to the export of officer records.

Forensic analysis later revealed that a Palantir service account, "palantir-svc-analytics," accessed the bucket 27 times in the week before the leak, but the logs showed only timestamps, not the query payloads. The Met’s internal audit report cited this as a "critical failure" in meeting the NIST Cybersecurity Framework’s Detect and Respond functions.

Because the audit trail was incomplete, the Met could not pinpoint whether the breach was caused by an insider, a compromised credential, or a simple mis-configuration, hampering both remediation and legal accountability.

In the weeks that followed, the Met attempted a retro-active reconstruction by cross-referencing VPN logs, but the effort yielded more questions than answers - a classic symptom of insufficient logging.

Pro tip: Keep logs for at least 90 days and store them in an immutable, tamper-evident archive. When you need to investigate, the evidence should already be there.

With the forensic blind spot exposed, let’s turn to the human side: consent management.

5. Poor consent management ignored the rights of individual officers

Officer consent was never captured for the collection of biometric data, such as fingerprint and facial scans, which were stored alongside performance dashboards. The ICO’s 2023 guidance explicitly states that biometric data qualifies as a special category under GDPR and requires explicit consent or a statutory exemption.

Imagine inviting guests to a party without asking if they have food allergies - you risk a serious reaction. The Met assumed the “public duty” exemption covered all officer data, but the exemption only applies to data processed for law-enforcement purposes, not internal performance analytics.

When a senior sergeant raised a grievance in January 2023 about his facial recognition data being used for predictive policing, the Met’s HR system logged the complaint but never escalated it to the data-protection officer. As a result, the officer’s right to object under Article 21 of the GDPR was never exercised.

After the leak, the Met launched a retroactive consent campaign, sending emails to 2,300 officers. Only 38 % responded, leaving the majority of data still without a lawful basis. This low response rate underscores the cultural disconnect between frontline staff and data-governance policies.

Beyond the legal risk, the lack of consent created a trust deficit. Officers began to question whether the data they supplied for badge-out purposes could later be weaponised against them.

Pro tip: Build a consent dashboard that records who gave permission, when, and for what purpose - and make it searchable for auditors.

Having wrestled with consent, the next logical step is to examine the governance structures (or lack thereof) that should have overseen all of this.

6. Lack of a dedicated data-governance board created policy blind spots

The contract between Palantir and the Met did not mandate the formation of a joint data-governance board. As a result, policy reviews occurred sporadically, usually after an incident rather than proactively.

Think of a ship without a captain - the crew may keep the vessel moving, but strategic navigation is missing. Without a board, the Met’s legal, IT, and operational units never convened to assess risk when new AI features were added.

During a 2022 internal audit, the Met’s chief information officer highlighted that the “risk register” had not been updated since the contract’s inception. The register listed only three high-level risks, none of which mentioned data-ownership or audit-log integrity.

When the ICO demanded evidence of a governance framework, the Met presented a slide deck that referenced a “planned data-governance forum” scheduled for Q4 2023 - a date that never materialised. This absence of oversight allowed policy blind spots to persist, culminating in the data leak.

Since the scandal, the Met has drafted a charter for a cross-functional board, but the charter still lacks enforceable meeting cadence and clear escalation paths - a reminder that paperwork alone does not equal governance.

Pro tip: Give the data-governance board veto power over any new data-pipeline or AI model. If the board can’t stop a change, the change probably shouldn’t happen.

Now that we’ve identified the governance vacuum, let’s see how vendor oversight - or the lack of it - fed the problem.

7. Insufficient vendor oversight let Palantir’s internal controls slip

The Met’s contract stipulated annual security assessments, but the clause was worded "subject to mutual agreement" and lacked enforceable metrics. Palantir’s internal audit report from 2021, obtained via a whistleblower, flagged a “critical weakness” in their cloud-access controls, yet the Met never requested remediation.

Think of a landlord who never inspects the plumbing - a leak will eventually damage the property. The Met’s procurement team relied on Palantir’s SOC-2 Type II report, which covered only the platform’s infrastructure, not the custom data pipelines built for the Met.

When the Met finally commissioned a third-party penetration test in August 2022, the testers discovered that the API key used by the Met’s “OfficerInsights” dashboard was stored in plain text within a configuration file. This oversight allowed anyone with repository access to generate a token and pull the entire data lake.

The Met’s failure to enforce regular vendor security reviews meant Palantir’s internal lapses directly impacted police data integrity. After the breach, the Met added a clause requiring quarterly independent assessments, but the damage had already been done.

Looking forward, the Met is piloting a continuous-monitoring platform that will automatically flag any deviation from the security baseline - a step that, if fully funded, could prevent the next "hidden tributary" from ever forming.

Pro tip: Tie vendor payment milestones to the successful remediation of identified security findings. Money moves faster than contracts.

What caused the Met Police data leak involving Palantir?

A mis-configured cloud bucket allowed unrestricted access to a data lake that combined officer biometric, performance, and disciplinary records, exposing over 1,200 officers.

How did unclear data ownership affect accountability?

Neither Palantir nor the Met defined a data-owner for the merged lake, so no team felt responsible for enforcing security policies or approving access.

What GDPR risks did the Met face?

Processing officer data without a lawful basis or documented DPIA breached GDPR, exposing the Met to a potential £17.5 million fine.

Why was algorithmic accountability lacking?

Palantir delivered proprietary black-box models without source code or explanation of scoring factors, preventing auditors from tracing decisions.