This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
The Chicago Police Department's 'Strategic Subject List' (SSL) was a predictive policing algorithm used to assign risk scores to individuals for potential involvement in gun violence. An independent audit by the RAND Corporation found the tool ineffective at reducing violence and noted that it was used for targeting arrests rather than social service intervention. The program faced significant criticism for racial bias, lack of transparency, and potential Fourth Amendment violations before being decommissioned in 2018.
Chicago Police Department (CPD)'s Strategic Subject List as output of an algorithm purportedly to identify victims or perpetrators of violence was reportedly ineffective, easily abused, and biased against low-income communities of color.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Unequal treatment of individuals or groups by AI, often based on race, gender, or other sensitive characteristics, resulting in unfair outcomes and unfair representation of those groups.
Also classified as
Police officers relied on the 'scientific' authority of the algorithm to justify targeted stops and arrests, laundering subjective biases into objective judgments.
The system failed to perform effectively or reduce crime rates, as confirmed by independent RAND Corporation audits.
The SSL algorithm systematically assigned higher risk scores to Black and low-income residents, resulting in unfair profiling and discriminatory policing outcomes for these groups.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The systemic racial profiling and feedback loops were unintentional and unexpected outcomes of deploying the predictive policing algorithm to allocate resources.
Added categories peak at 4: Severe (Psychological Harm, Inferred)
The Chicago Police Department deployed a predictive policing algorithm (SSL) that generated biased risk scores for 400,000 residents. While causing significant local civil rights concerns and systemic racial profiling, the incident had negligible direct impact on national security and was ultimately resolved through local oversight and decommissioning.
The SSL algorithm acts as an Assistant by analyzing arrest and victimization data to generate risk scores and suggestions for police intervention, requiring human officers to make the actual decisions on the ground.
High Risk: The report describes an AI system used in law enforcement for profiling and predicting crimes, which has significant implications for safety, fundamental rights, and public interests.
The primary goal of the SSL system was to predict which individuals were most likely to be involved in gun violence, either as victims or perpetrators, to help police allocate resources and conduct targeted interventions.
The reports provide extensive, detailed, and consistent accounts of the Chicago Police Department's Strategic Subject List (SSL), its development, deployment, and subsequent decommissioning. Independent audits by the RAND Corporation and investigative journalism provide empirical data on the system's scale (nearly 400,000 people scored) and its disproportionate impact on Black residents (56% of young Black men scored). The technical inputs and limitations of the algorithm are well-documented, leaving little ambiguity about the core facts of the incident.
None
Technology
Feedback Loop Amplification
Algorithms reinforce prior biases by predicting crime where arrests occurred.
Proprietary Black-Box Algorithms
Trade secrets prevent independent review of how variables are weighted.
Inaccurate Prediction Models
Models like SSL had little to no impact on reducing actual crime rates.
Data Inputs
Biased Historical Arrest Data
Models train on arrests reflecting past policing bias rather than actual crime.
Use of Unconvicted Arrest Records
Systems use arrest records instead of convictions to calculate threat scores.
Incomplete Crime Reporting
Less than half of violent crimes are reported, biasing the input data.
Human Factors
Overreliance on Algorithm Outputs
Officers trust scores without understanding the underlying biases or weights.
Lack of Officer Training
Field officers receive little training on how to interpret and use lists.
Confirmation Bias in the Field
Officers patrol targeted areas expecting crime, confirming prior beliefs.
Process and Methods
Lack of Public Transparency
Algorithms are deployed without public input or explanation of variables.
Inadequate Intervention Protocols
Departments fail to pair risk lists with effective social interventions.
Absence of Independent Audits
Very few independent evaluations of predictive policing tools are conducted.
Management
Prioritizing Tech over Strategy
Management adopts software as a panacea without clear operational plans.
Failure to Release Internal Audits
Leadership keeps efficacy reports and algorithmic data shielded from review.
Decommissioning without Long-term Plans
Programs like SSL are shut down due to poor planning and low efficacy.
Regulatory Environment
Weak Fourth Amendment Protections
Predictive scores are used to bypass reasonable suspicion requirements.
Absence of Algorithmic Standards
No regulatory standards govern the use of big data in law enforcement.
Lack of Public Accountability
Legal barriers prevent citizens from challenging their inclusion on lists.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
