# Switzerland - PreCOBS predictive policing with questionable efficacy
Switzerland, 2020-2021 | Severity: N2

## What Happened
Several Swiss cantonal police departments deployed an automated predictive policing system called "PreCOBS" (Pre-Crime Observation System) to predict and prevent domestic burglaries. The algorithm analyzed historical crime data, geographical locations, and timing to output "risk zones" where burglaries were predicted to occur.
The use of PreCOBS raised questions regarding predictive validity, operational effectiveness, transparency, and the possibility that historical policing patterns could influence subsequent patrol allocation.
Independent assessments and public reporting questioned whether the system produced sufficiently strong evidence of improved policing outcomes compared with conventional methods.

## How MEG 1 acted
MEG 1 would treat PreCOBS as a high-impact decision-support system whose outputs could influence the geographic allocation of police attention.
The system would require continuous validation against real-world outcomes rather than acceptance of predictive scores as inherently authoritative.
DAI monitoring would test whether generated risk zones produced stable, reproducible, and meaningfully superior predictive performance.
ISR controls would require examination of whether repeated use of historical police data created self-reinforcing feedback loops in which previously patrolled areas generated more recorded incidents and therefore attracted further algorithmic attention.
Predictive output would remain advisory and could not independently authorize enforcement activity.

## MEG 2 Analysis - Legal Framework
* **Agent level:** N2 (management) - bounded operational decision-support system capable of influencing public-resource allocation and policing priorities without independently exercising coercive authority.
* **Cause (6.1):**
  * **(a) system defect** - a predictive model whose operational value and effectiveness could not be independently demonstrated, while historical crime patterns risked being recursively translated into future policing priorities.
  * **(b) automated decision-support error** - risk-zone predictions could influence patrol allocation despite uncertain predictive value and limited evidence that the system produced better outcomes than conventional policing methods.
* **Attachment of liability:** Under the MEG 2 allocation model, primary accountability would attach to the public authority using predictive outputs to allocate policing resources. Developer accountability would depend on demonstrated defects, misleading performance claims, or failures in the design and validation of the predictive system.

## Procedural Mechanism
* **6.2 (architectural human confirmation):** Predictive risk zones must remain advisory. Material changes in patrol allocation or enforcement activity based on algorithmic predictions must require documented human assessment rather than automatic execution.
* **7.1 (independent forensics):** Input crime data, generated risk zones, resulting deployment decisions, and subsequent outcomes must be independently logged so that the system's real-world effectiveness and potential feedback effects can be audited.
* **9.4 (least privilege):** Predictive outputs must not independently trigger enforcement actions or acquire operational authority beyond narrowly defined decision-support functions.