Flock Mandates Case Numbers and Automated Audits After Police Camera Misuse Allegations

01Flock Makes Case Numbers and Automated Audits Mandatory for Police Searches After License-Plate Surveillance Abuse Controversy

Flock Safety, a police-technology company operating about 120,000 license-plate-reading cameras, is tightening access to the nationwide network that police departments use to search vehicle-location data. The move follows allegations that officers used the system for unauthorized purposes, including tracking and harassing current or former romantic partners.

A recent Washington Post investigation identified 46 cases in which officers were accused of misusing Flock cameras. After mounting protests and lost municipal contracts, Flock announced that two safeguards introduced as optional settings last year—case-number entry and automated auditing—will now be mandatory.

Officers will have to enter a criminal case number before conducting a search. The requirement is intended to connect each query to a legitimate investigation, but Flock will not verify the numbers. That leaves room for officers to enter false or irrelevant information, much as they circumvented an earlier prompt asking them to state a search’s purpose. The ACLU found entries including the generic term “investigation”; officers at one Oregon department entered “hehehe” 20 times.

Flock will also require agencies to use an automated auditing system that analyzes officers’ search activity and flags suspicious behavior to administrators. The company has not disclosed the tool’s accuracy or opened it to independent evaluation, so there is no public evidence yet showing how reliably it identifies misuse.

Other changes address who can search camera data and how long it remains available. Departments can restrict outside agencies’ access according to their declared purpose—for example, permitting searches related to kidnapping but excluding immigration enforcement. That protection still depends on officers accurately describing why they are searching. Flock is also recommending that agencies shorten data retention from 30 days to seven, although departments may reject that advice.

The backlash has already cost Flock business. NPR reported in February that at least 30 cities had canceled contracts during the previous year, while an activist group puts the total higher; the precise number is unknown. Flock says the cancellations represent only a small share of the roughly 5,000 agencies under contract, and CEO Garrett Langley has attributed the company’s customer losses mainly to misinformation.

The new requirements mark a shift from offering safeguards to enforcing them, but whether they reduce abuse remains unknown. An ACLU policy counsel said judging their effectiveness would require independent researchers to examine Flock’s systems.

02Google Launches Gemini 3.7 Flash Three Weeks After Its Last Update, Halving 3.6 Flash’s Original Price Through Year-End

Google has released Gemini 3.7 Flash, the latest version of its model series for high-volume workloads, just three weeks after launching Gemini 3.6 Flash. The company is positioning the new model as its primary option for coding and AI agents—systems that can plan tasks, use tools and take actions for users.

Google says developer feedback and algorithmic changes produced gains in software engineering, knowledge work and web development. On coding evaluations published by the company, 3.7 Flash scored 43.6% on FrontierCode 1.1 Main, compared with 34.4% for 3.6 Flash. Its DeepSWE v1.1 score rose from 49.0% to 65.3%.

The company also reported improvements beyond coding. On GDP.pdf, an evaluation of models’ ability to process complex documents, 3.7 Flash scored 34.0%, against 22.0% for its predecessor. On AutomationBench, which tests business workflows, the scores were 30.4% and 17.0%, respectively. For web development, 3.7 Flash recorded an Elo score of 1,588 on Arena.ai’s WebDev Arena, versus 1,538 for 3.6 Flash.

Google says the model is more likely to clarify intent, follow instructions and adapt when it encounters obstacles. It also claims more disciplined multistep planning and tool use can reduce retries and manual oversight, although it has not quantified those effects in production workflows. The published results do not include independent testing.

Through the end of the year, developers will pay an introductory price of $0.75 per million input tokens and $3.75 per million output tokens. Google describes that as half the original price of 3.6 Flash; it has not disclosed what 3.7 Flash will cost after the promotion ends.

Gemini Spark is receiving the model immediately. Spark is Google’s continuously running personal agent for Google AI Pro and Ultra subscribers in more than 160 countries. Google says 3.7 Flash should improve Spark’s use of Google Workspace tools for tasks such as consolidating files, drafting emails and updating status documents.

Developers can access 3.7 Flash through the Gemini API, Google AI Studio, Android Studio and Google Antigravity. Enterprises can use it through the Gemini Enterprise Agent Platform and app, while eligible individual subscribers can access it through Spark in supported countries.

03CGTrader Says AI Models Make Up One in Six Recent Uploads but Just One-Ninetieth of Platform Revenue

AI-generated 3D models are arriving on CGTrader far faster than buyers are paying for them. The online marketplace, which offers more than two million models for videos, games and 3D printing, says about one in six recent uploads is AI-generated, while those assets produce only $1 of every $90 in platform revenue.

The finding comes from CGTrader marketplace data collected between June 2025 and May 2026 and surveys of buyers. It reveals a sharp mismatch between supply and demand on the platform: AI has lowered the barrier to producing and uploading models, but buyers still prioritize whether an asset works. CGTrader did not disclose absolute upload, sales or revenue totals for AI-generated and human-made assets, or the survey’s sample size and methodology.

Quality ranked above price as the most important purchasing factor among surveyed buyers. Of buyers who had purchased AI-generated assets, 20% said the quality was inadequate, 7% could use the asset only after extensive editing and 5% said it performed well.

The stakes are especially high in 3D printing, where a defective model can break or fail to print correctly. Only 4% of surveyed buyers in that category said AI-generated assets worked well, helping explain why rapid growth in available models has not translated into comparable revenue.

CGTrader is not opposed to using AI. The company has partnered with Tencent on an AI-assisted workflow that lets designers refine topology and textures, separate parts, create variations and prepare models for production. Chief executive Dalia Lašaitė said CGTrader wants AI to accelerate mechanical tasks while designers retain creative direction, rather than simply increase the number of AI-made assets.

Lašaitė also acknowledged that uploads are growing faster than purchases, potentially making human artists’ work harder to find. CGTrader plans to rank models using quality and performance signals—including commercial results and buyer ratings—instead of rewarding upload volume. Whether those changes will improve exposure for human creators remains unknown.