Case Study 33-2: Stop LAPD Spying Coalition and the Fight Against Predictive Policing

Community-Based Counter-Surveillance in Los Angeles

Background: The Los Angeles Surveillance Landscape

Los Angeles has, for much of its modern history, been a laboratory for surveillance policing. The LAPD's use of gang injunctions (civil orders restricting movement and association of people designated as gang members), its history of infiltrating political organizations, and its aggressive data collection practices have made it one of the most documented and contested police surveillance programs in the United States.

In the early 2010s, the LAPD adopted two surveillance technologies that drew particular scrutiny:

Suspicious Activity Reporting (SAR): Derived from post-9/11 counterterrorism intelligence frameworks, SAR required officers to report observations of "suspicious" behavior — photographing infrastructure, wearing backpacks, making notes in public, etc. — into a centralized database shared with federal agencies including the FBI. Critics documented that SAR categories were vague enough to encompass ordinary civil liberties activities, and that implementation was racially and politically biased.

PredPol (Predictive Policing): The Los Angeles Police Department adopted PredPol (later renamed Geolitica) beginning in 2011. PredPol uses historical crime data and algorithmic modeling to generate predictions of where crimes are likely to occur — producing maps that direct patrol resources toward predicted crime locations. The LAPD deployed it in several divisions and reported reductions in crime rates that it attributed partly to the program.

The Stop LAPD Spying Coalition: Origins and Methods

The Stop LAPD Spying Coalition was founded in 2011 by community organizers, civil liberties lawyers, and residents of Los Angeles communities most affected by LAPD surveillance. Its founder and longtime director, Hamid Khan, had a background in immigrant rights organizing and brought that movement's emphasis on community leadership (those most affected should lead the organizing) to surveillance accountability work.

The coalition's methodology distinguished it from many civil liberties organizations. Rather than primarily pursuing litigation (the ACLU model) or technical research (the Citizen Lab model), the coalition centered community-based data analysis and public education.

Public records strategy: The coalition filed extensive public records requests for LAPD surveillance data — SAR reports, PredPol deployment records, contracts with data companies, training materials. These requests often faced resistance and delays; the coalition developed expertise in navigating and challenging those delays.

Community data analysis: When records were obtained, the coalition convened community meetings to analyze them collectively. Residents of communities affected by SAR reporting reviewed the reports about their own neighborhoods. The analysis was not conducted by experts looking at communities from outside but by community members examining data about themselves.

Counter-mapping: The coalition produced visual analyses of SAR distribution and PredPol deployment zones, mapping the geography of surveillance against the geography of race and poverty in Los Angeles. These visual representations made the abstract concept of "biased surveillance" concrete: specific neighborhoods, specific percentages, specific patterns.

Public education: The coalition produced plain-language reports, held community workshops, and built relationships with journalists who could translate technical surveillance analysis into accessible news coverage.

The PredPol Challenge: Documented Bias and Feedback Loops

The coalition's most significant technical challenge was to the predictive policing program. PredPol's designers and LAPD defenders argued that the algorithm was race-neutral — it used crime report data, not race data, to generate predictions.

The coalition's response, developed in collaboration with academic researchers, identified the central flaw in this argument: feedback loops in predictive policing systems amplify historical bias rather than correcting for it.

The mechanism: 1. Historical crime data reflects not just where crime occurred but where police looked for crime — communities with higher patrol density have higher reported crime rates partly because there are more officers to report it 2. PredPol uses historical crime data to predict future crime locations 3. Police resources are directed toward predicted locations 4. Police find (and report) crime in areas where they look for it 5. This finding generates new crime data in those areas 6. The new data reinforces the original prediction 7. The cycle repeats, continuously reinforcing the concentration of patrol in the same neighborhoods

The result: communities that were over-policed historically were predicted to have high crime, which sent more police to those communities, which found and reported more crime, which reinforced the prediction. The algorithm did not introduce racial bias — it inherited and amplified existing bias.

This is the function creep problem discussed in earlier chapters, applied to predictive systems: tools designed to be objective inherit the bias of the historical data they learn from, then launder that bias through the authority of algorithmic neutrality.

The Policy Outcomes

The coalition's work contributed to significant policy changes:

2020 — LAPD ends PredPol: Following sustained pressure from the Stop LAPD Spying Coalition, investigative journalism, and public attention to police reform after the murder of George Floyd, the LAPD Board of Police Commissioners voted in 2020 to end the city's contracts with PredPol/Geolitica. Los Angeles became one of the first major cities to end a predictive policing program through public political pressure rather than legal challenge.

2019 — Suspicious Activity Reporting reforms: Following coalition advocacy, the LAPD implemented narrower definitions of "suspicious activity" and added oversight requirements. The coalition argued these reforms were insufficient — they changed policy on paper without changing police culture — but they represented measurable policy response.

Citywide surveillance ordinance: Los Angeles adopted a surveillance technology ordinance (STO) in 2019 requiring city council approval before LAPD could deploy new surveillance technologies and requiring public reporting on existing technologies. Similar ordinances have been adopted in Oakland, San Francisco, Boston, and other cities, often in connection with campaigns against facial recognition.

The Coalition's Broader Significance

The Stop LAPD Spying Coalition's significance extends beyond Los Angeles. It demonstrated that community-based organizing — not just litigation or technical research — could produce meaningful accountability for surveillance programs. Its methodology (community data analysis + policy advocacy + public education) has been adopted by organizations in other cities facing similar surveillance challenges.

Its theoretical contribution was also significant: the coalition framed predictive policing not as a technical failure (the algorithm is wrong) but as a manifestation of structural racism (the algorithm accurately reflects and amplifies the racially biased distribution of police attention). This reframing shifted the debate from "fix the algorithm" to "question the premises of predictive policing."

This distinction matters. A "fix the algorithm" response accepts the premise that more accurate prediction is the goal and focuses on correcting technical errors. A "question the premises" response asks whether prediction-based policing is appropriate regardless of its accuracy — whether concentrating police resources in communities based on statistical models is compatible with constitutional rights and human dignity, even if the models are technically sound.

The coalition's position was abolitionist rather than reformist: predictive policing was not a good idea implemented badly but a bad idea that should be ended, not improved.

Analysis Questions

1. The Stop LAPD Spying Coalition used community data analysis — residents examining data about their own communities — rather than expert analysis of community data. What are the advantages of this approach? What might be its limitations? When is expert analysis necessary, and when does it substitute for rather than support community leadership?

2. PredPol's defenders argued that the algorithm used crime data, not race data, and was therefore race-neutral. The coalition's response was that algorithms can amplify racial bias even without using race as a variable. Evaluate this debate. Does the coalition's argument mean that any use of historical crime data in policing is necessarily biased?

3. The coalition's position was abolitionist — end predictive policing, not reform it. Other organizations argued for reforms (better oversight, bias audits, algorithmic transparency). On what grounds might one choose abolition over reform as a political strategy? What risks does each approach involve?

4. The coalition's success in ending PredPol in 2020 coincided with the broader George Floyd moment and national debate about policing. Is this a story of successful advocacy, or of fortuitous political timing? What does the answer imply for how surveillance activists should think about strategy?

5. Los Angeles's surveillance technology ordinance requires city council approval for new surveillance technologies. Evaluate this governance mechanism: What does it accomplish? What are its limitations? How might similar mechanisms be improved?

6. The coalition frames surveillance as a racial justice issue rather than (only) a privacy issue. How does this reframing change who participates in surveillance politics, what demands seem reasonable, and what outcomes count as success?

This case study connects to Chapter 33 Section 33.8 (movement-level activism) and backward to themes of social sorting and function creep. It connects forward to Chapter 35 (facial recognition and its bias) and Chapter 38 (predictive AI in criminal justice).