Gizmodo analysis finds PredPol predictions targeted Black, Latino, and low-income areas
Independent analysis of PredPol prediction logs found the software repeatedly generated predictions concentrated in Black, Latino, and lower-income neighborhoods. The findings, reported by Gizmodo/The Markup and discussed in multiple news outlets, showed patterns consistent with bias arising from the model's training data and operational use.
Analysis found PredPol's predictions were concentrated in Black, Latino, and low-income neighborhoods because the model mirrored biased historical crime-report data.
Key facts
- What
- Independent analysis of PredPol prediction logs found the software repeatedly generated predictions concentrated in Black, Latino, and lower-income neighborhoods.
- Incident date
- Dec 2, 2021
- Who
- PredPol (now Geolitica)
- Failure mode
- Policy Violation
- AI surface
- Algorithmic Decision
- Severity
- High
What happened
Journalistic analysis (Gizmodo/The Markup) obtained millions of PredPol prediction records and reported that neighborhoods with higher Black, Latino, and low-income populations received far more crime predictions between 2018 and 2021. The analysis found that PredPol predictions often clustered in the same neighborhoods repeatedly and that wealthier, whiter neighborhoods were predicted far less frequently. Some police departments later stopped using the software and PredPol (now Geolitica) disputed parts of the analysis while acknowledging the files appeared to be generated by the company.
What broke inside the model
- 01 · TriggerA prompt pushes against a deployment boundary.
- 02 · Model stepThe model produces the disallowed output.
- 03 · Control gapNo enforcement blocks it at generation time.
- 04 · FailureThe output crosses the policy line.
- 05 · ConsequenceA limit the business set is breached in public.
The output crosses a policy boundary the deployment had defined.
The system relied on historical police-reported crime data and produced location-based risk boxes; this input reflected unequal reporting and enforcement patterns and acted as a proxy for race and poverty. The algorithmic design and operational guidance to 'get in the box' created a feedback loop where increased patrol presence could reinforce the historical signals the model used, concentrating predictions and policing in already overpoliced communities.
What it cost
Sources
- PressCrime Prediction Software Promised to Be Free of Biases. New Data Shows It Perpetuates Themgizmodo.com
- PressPolice are using software to predict crime. Is it a ‘holy grail’ or biased against minorities?washingtonpost.com
- PrimaryIncident 54: Predictive Policing Biases of PredPolincidentdatabase.ai
Cite this entry
https://failureindex.ai/failures/predpol-predictions-disproportionately-targeted-black-lAI Failure Index. "Gizmodo analysis finds PredPol predictions targeted Black, Latino, and low-income areas" (FI-0446). Realm Labs. https://failureindex.ai/failures/predpol-predictions-disproportionately-targeted-black-l (indexed Jun 10, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0446. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm fits
- Prism
- OmniGuard
This entry sits in the index's predictive wing: a system that scores, ranks, perceives, or steers rather than generates. Realm's runtime layer is built for the generative and agentic systems now moving into these same decision seats, where it watches a model's internal state and holds an unsupported claim or an unchecked action before it commits. The control gap on this record, an automated decision that reached people with no runtime check in front of it, is the same gap. The index keeps predictive failures on the record because the pattern carries straight into the systems shipping today.