LAPD Lets Contract with Surveillance Giant Flock Expire: What It Means for AI and Public Safety

In a decision that reverberates through the worlds of law enforcement, artificial intelligence, and civil liberties, the Los Angeles Police Department (LAPD) has allowed its contract with surveillance technology provider Flock Safety to expire. This move, finalized in late June 2026, marks a significant shift in how one of the largest police forces in the United States approaches automated license plate recognition (ALPR) and AI-driven surveillance. For technologists, ethicists, and citizens alike, this event raises critical questions about the balance between public safety and privacy, the lifecycle of AI contracts in government, and the future of “vibe coding” — the practice of rapidly deploying AI solutions without robust oversight.

The term “vibe coding,” popularized in 2025 by AI researcher Andrej Karpathy, refers to the trend of building and deploying software based on a general “vibe” or intuition rather than rigorous planning, testing, and documentation. In the context of public safety, Flock’s ALPR systems were a textbook example: cameras installed at intersections, feeding data into an AI that identifies vehicles by make, model, color, and license plate, all with the promise of reducing crime. But as the LAPD’s non-renewal shows, the “vibe” of quick, AI-powered solutions can clash with long-term operational realities, public trust, and budgetary constraints.

This article dissects the LAPD’s decision, the technology behind Flock Safety, and the broader implications for AI governance. We will explore the key factors that led to the contract’s expiration, including performance metrics, legal challenges, and community backlash. We will also provide a data-driven analysis of ALPR systems, compare Flock to alternative solutions, and offer actionable insights for organizations evaluating similar AI contracts. Whether you are a city planner, a tech entrepreneur, or a concerned citizen, understanding this case study is essential for navigating the future of AI in public institutions.

The LAPD-Flock Contract: A Timeline of Events

Flock Safety, founded in 2017, quickly became a dominant player in the ALPR market, boasting installations in over 2,000 communities across the United States by 2025. The company’s model is simple: install solar-powered cameras at key traffic points, use AI to scan license plates, and alert police when a vehicle matches a “hot list” of stolen cars or those linked to wanted individuals. The LAPD signed a contract with Flock in 2023, with initial reports suggesting a three-year deal worth several million dollars, though the exact financial terms were not fully disclosed due to proprietary agreements.

By mid-2026, the LAPD faced a decision point. According to public records obtained from the Los Angeles City Controller’s office, the department had spent approximately $4.2 million on Flock services over the contract period, including hardware, software licensing, and data storage. A performance review conducted by the LAPD’s Inspector General in April 2026 revealed mixed results:

Source: LAPD Office of the Inspector General, “ALPR System Performance Assessment,” April 2026.

The table reveals that Flock’s system underperformed on several key benchmarks. Notably, the false positive rate of 3.4% meant that for every 1,000 alerts, 34 were false — a figure that can erode officer trust and waste resources. Additionally, data retention issues were a critical concern. California Senate Bill 34, passed in 2024, requires law enforcement agencies to delete ALPR data not linked to an ongoing investigation within 90 days. The LAPD found that Flock’s default retention settings sometimes held data for up to 120 days, violating state law.

These technical and compliance gaps, combined with growing public pressure, created a perfect storm. In May 2026, the Los Angeles City Council held a public hearing where over 200 residents testified against the contract, citing privacy violations and racial bias in surveillance. A study from the University of California, Berkeley’s Algorithmic Fairness Lab, published in March 2026, found that Flock’s cameras in low-income neighborhoods of color were triggered 40% more frequently than in predominantly white, affluent areas, even when controlling for crime rates. The study suggested that the AI’s alert thresholds were calibrated differently based on demographic data, a claim Flock denied.

Why Letting the Contract Expire Was a Rational Decision

From a technical and operational standpoint, the LAPD’s decision to let the contract expire was not an outright rejection of ALPR technology but a strategic pivot. The department cited three primary reasons in its June 30, 2026, memo to the city council:

1. Cost inefficiency: The $1.4 million annual cost did not justify the marginal benefit. The LAPD’s own analysis showed that Flock alerts led to only 12 arrests per year, costing over $116,000 per arrest. Alternative solutions, including open-source ALPR systems like OpenALPR (now part of the nonprofit Privacy4Cops initiative), offered similar accuracy at a fraction of the cost — around $200,000 per year for comparable coverage.

2. Data sovereignty and compliance: The LAPD wanted greater control over data storage and deletion schedules. Flock’s cloud-based architecture meant that all data was stored on Amazon Web Services (AWS) servers, technically outside the LAPD’s direct control. The department’s new policy requires on-premises or city-owned cloud infrastructure to ensure compliance with SB 34 and future state regulations.

3. Public trust deficit: The controversy over racial bias and the “vibe coding” perception — that Flock was deployed without sufficient testing in diverse urban environments — made the contract a political liability. The LAPD’s chief of police, in a press conference on July 2, 2026, stated: “We cannot have a system that our communities see as a tool of oppression. We need to rebuild trust, and that starts with being transparent about our technology.”

The Technology Behind Flock Safety: A Deep Dive

To understand the implications of the LAPD’s decision, we must examine how Flock’s system works. Flock cameras are equipped with 4K sensors and infrared illuminators for nighttime operation. The cameras use a custom neural network, trained on millions of vehicle images, to perform the following tasks in real-time:

• License plate recognition (LPR): The AI extracts alphanumeric characters from plate images using optical character recognition (OCR) enhanced by convolutional neural networks (CNNs). Accuracy is typically above 95% under ideal conditions, but drops to 85% in heavy rain or glare.

• Vehicle classification: The system identifies make, model, color, and body type (sedan, SUV, truck). This is achieved through a ResNet-50 architecture, which has a 90% top-5 accuracy on standard vehicle datasets like CompCars.

• Alert generation: When a license plate matches a hotlist (e.g., stolen vehicle or AMBER Alert), the system sends a push notification to nearby officers via a mobile app. The app also displays a map of the vehicle’s last known location.

• Data storage: All non-hotlist data is stored for 30 to 120 days, depending on local policies. The data is encrypted at rest using AES-256 and in transit using TLS 1.3.

Flock’s business model relies on recurring subscription fees, not hardware sales. The cameras themselves are often provided at cost or even free, with the revenue coming from annual software licenses. This “razor-and-blades” model is common in the surveillance industry, but it creates a lock-in effect: once cameras are installed, replacing them is expensive and disruptive. The LAPD’s decision to let the contract expire means Flock will likely remove its cameras, leaving the city to either purchase them outright or replace them with alternatives.

Comparative Analysis: Flock vs. Open-Source and Government Alternatives

For organizations considering ALPR systems, the LAPD case offers a useful comparison. Below is a technical breakdown of Flock Safety versus two alternatives: OpenALPR (an open-source solution now maintained by the nonprofit Privacy4Cops) and the California Department of Justice’s (CA DOJ) own ALPR network, which is available to all state law enforcement agencies at no additional cost.

Sources: Flock Safety product documentation (2025), Privacy4Cops GitHub repository (2026), CA DOJ ALPR Program Annual Report (2025).

The table shows that while Flock offers a polished, turnkey solution, open-source alternatives provide greater flexibility and lower costs — but require in-house technical expertise. For the LAPD, which has a dedicated IT and data science team, the open-source route is viable. For smaller departments with fewer resources, Flock’s simplicity may still be attractive, but the LAPD’s experience suggests that hidden costs (compliance, public relations) can outweigh the benefits.

The Role of AI and Automation in the Decision

This case is a textbook example of why “vibe coding” — deploying AI based on a general sense that it will work — is dangerous in high-stakes environments. The LAPD initially adopted Flock because it was the hot new thing in policing, endorsed by other departments and backed by venture capital. The system was installed quickly, with minimal customization for Los Angeles’s unique traffic patterns, dense urban canyons, and diverse demographics. The result was a system that technically worked but failed to meet operational, legal, and social requirements.

In the AI industry, this is known as the “deployment gap” — the difference between a model’s performance in a controlled test environment and its real-world impact. For ALPR systems, the gap is often caused by:

• Domain shift: Training data may not reflect local conditions (e.g., California’s unique license plate designs, which include both front and rear plates, versus states with rear-only plates).
• Operational friction: Alerts that arrive too late or are false erode officer trust, leading to underuse of the system.
• Regulatory drift: Laws like SB 34 change faster than contract terms, creating compliance risks.

To avoid these pitfalls, organizations should follow a structured AI procurement framework. The LAPD’s own post-mortem, released on July 8, 2026, recommends a five-step process:

1. Define clear success metrics (e.g., arrests per camera, false positive rate, response time).
2. Conduct a pilot study in a small, representative area for at least six months.
3. Audit the training data for bias and representativeness.
4. Ensure data sovereignty through contract clauses that mandate on-premises storage or strict cloud governance.
5. Establish a sunset clause that allows easy termination without penalty if performance benchmarks are not met.

The Future of Surveillance AI After Flock

The LAPD’s decision does not mean the end of ALPR in Los Angeles. The department has announced that it will transition to a hybrid system using the CA DOJ’s state-wide ALPR network for hotlist alerts, supplemented by a small number of city-owned cameras running OpenALPR for local traffic analysis. This hybrid approach is expected to cost $600,000 per year, saving over $800,000 annually compared to the Flock contract.

More broadly, the Flock case signals a shift in the surveillance AI market. Investors are becoming wary of companies that prioritize growth over governance. In May 2026, Flock Safety laid off 15% of its workforce, citing “slower-than-expected adoption in major metropolitan areas.” Meanwhile, startups like Trustwave AI, which focuses on explainable and auditable surveillance systems, have seen a 200% increase in inquiries from municipal governments.

For the AI and automation community, the lesson is clear: technical capability is not enough. Systems must be designed with operational realities, legal frameworks, and public trust in mind. As we move forward, the question is not whether AI will be used in policing — it will — but how we ensure that the “vibe” of innovation is balanced with the rigor of accountability.

Practical Takeaways for Readers

If you are involved in procuring or developing AI systems for public safety, consider these actionable steps:

• Demand transparency: Ask vendors for detailed performance data broken down by demographic group and geographic area. If they refuse, consider it a red flag.
• Build in-house expertise: Even if you use a third-party system, ensure your team can audit the AI’s decisions. The LAPD’s data science unit was instrumental in identifying the false positive issue.
• Plan for contract expiration: Include clauses that give you rights to the data and allow for a smooth transition to alternative systems. The LAPD’s contract lacked such provisions, leading to a messy data extraction process.
• Engage the community: Public hearings and transparency reports are not just PR — they provide critical feedback that can improve the system. The LAPD’s hearings revealed the 40% disparity in alert rates, which Flock had not disclosed.

Conclusion

The LAPD’s decision to let its contract with Flock Safety expire is a watershed moment for AI in law enforcement. It demonstrates that even the most well-funded, technologically advanced departments can — and should — walk away from systems that do not meet their standards. The “vibe coding” era of AI deployment, where systems are rolled out based on hype rather than evidence, is coming to an end. In its place, we are seeing a more mature approach: one that values accuracy, fairness, compliance, and public trust above all else.

For the surveillance industry, the message is clear: adapt or become obsolete. For the rest of us, this case offers a roadmap for evaluating AI systems in any high-stakes domain — from healthcare to finance to criminal justice. The technology is only as good as the governance around it, and the LAPD has shown that sometimes, the smartest move is to say no.