A City Contract Can Turn Your Public Posts Into a Police File

The revealing object is a line item in New York City’s procurement system: access to software from Voyager Labs, an Israeli company that sold tools for analyzing social-media activity. The NYPD contract, awarded in 2018 with a ceiling in the high single-digit millions, did not look like a new police power. It looked like enterprise software.

That administrative disguise is the first useful fact. A city does not need to announce a social-media surveillance program when it can purchase a subscription, classify it as an analytical service and let the vendor define what analysis means. There is no ceremonial installation. No camera appears above the block.

People keep posting about protests, landlords, school fights, local politics and police encounters, unaware that public speech can now be sorted inside a system built for investigators rather than followers.

The Voyager line item bought more than a better search bar. Company materials described software that could collect public information, resolve identities across accounts and display connections among people, places and topics. Reporting on the NYPD contract and later litigation involving Meta showed the broader business model: gather social data at scale, structure it and sell the resulting visibility to institutions.

The city pays. The vendor gets recurring revenue. The platforms and their users supply the raw material.

The post becomes an investigative object

A person searching Instagram sees accounts, captions and whatever the platform’s ranking system chooses to place nearby. Monitoring software approaches the same material as a dataset. Investigators can search terms, account names and locations, then move outward through reposts, mentions, shared links and visible connections.

That distinction sounds technical until the search lands on you. A post saying that police have blocked a street was written for friends or neighbors. Inside an analytical platform, it can become a lead attached to a location and a network of accounts that interacted with it. The speech was never addressed to government.

Public availability supplies access anyway.

A social graph, a map of observable connections among accounts, gives the software its institutional value. Police already know how to search a name. Vendors promise to show the surrounding pattern: who appears repeatedly, which accounts amplify the same material, where profiles overlap and which identity may sit behind several usernames. Those links can be weak, ironic or accidental.

The interface still renders them with the visual authority of a case file.

Voyager marketed forms of identity resolution and relationship analysis, meaning the system attempted to connect fragments of online activity to a person or group. Such an inference is not a discovered fact. It is a vendor’s conclusion, produced from available signals and packaged for someone with state power.

This is where the Voyager procurement line matters. The city did not merely acquire posts that anyone could read. It paid to remove the friction that ordinarily limits reading: the hours required to search, compare profiles, preserve material and sketch relationships by hand. Surveillance at municipal scale is often an efficiency purchase.

The machine makes looking cheaper, so the institution looks more often.

Procurement writes the practical rules

City purchasing records can reveal the vendor, contract term and spending authority while saying little about operational limits. A description such as social-media analytics does not tell residents which units may run searches, what level of suspicion is required or whether a protest slogan can become a monitoring term.

The gap works for both parties. Vendors can advertise broad capability without accepting responsibility for an agency’s query. Agencies can say the tool searches public information without explaining how aggregation changes the character of that information. Each points to the other when scrutiny arrives.

Public information is the phrase doing the laundering. A speech at a public meeting is public too, but the government does not ordinarily receive an indexed archive of everyone who nodded, repeated a phrase elsewhere and followed the speaker months later. Software can preserve those contextual traces and make them available in one interface, which changes the cost and reach of observation even when each underlying post was technically visible.

Retention deepens the problem. A platform may delete a post, suspend an account or make older material difficult to find. A monitoring product may allow an investigator to save search results, export material or place screenshots and reports into another police system. The vendor’s database is therefore only one layer.

Intelligence can survive in case-management tools, email attachments and records produced for an investigation.

The contract rarely settles all of this. Agency policy, technical settings and vendor architecture divide control, while public-records law adds another layer that varies by jurisdiction. A city may impose retention limits for the monitoring platform yet preserve an exported report under a separate schedule. Deletion from one screen does not guarantee deletion from government custody.

That Voyager line item purchased access for a defined term. The information officers derived from it could have a much longer administrative life.

The inference is the product

Keyword monitoring is easy to understand. Inference deserves more attention because it turns ambiguous activity into institutional suspicion.

Vendor systems may classify sentiment, identify recurring associations or suggest that separate profiles belong to the same person. Sentiment analysis, automated labeling of language as favorable, hostile or neutral, performs badly with slang, sarcasm and context. Police work supplies no magic correction. An investigator can accept the label, override it or treat it as a reason to keep looking.

Network analysis carries a similar problem. Following an account can indicate friendship, opposition, professional interest or nothing beyond one impulsive tap. Shared hashtags may reflect coordination, but they also organize mass public conversation. A repost can be endorsement or criticism.

The software compresses these distinctions because its commercial purpose is to produce usable patterns, not to preserve every ambiguity that makes online speech human.

The vendor’s incentive is plain. A product that returns the same public posts available through an ordinary browser is difficult to sell for serious money. It must promise hidden structure: identities joined across services, relationships surfaced from interaction and emerging activity identified before officials would notice it themselves.

Police departments have a matching incentive. A system that produces leads without requiring an officer to begin with a known suspect expands the field of possible attention. Agencies can monitor an event, issue or neighborhood, then discover people through the search. Suspicion arrives after collection rather than before it.

Public reporting by the Brennan Center for Justice and the ACLU has shown this pattern across agencies and vendors, including products from Geofeedia, Media Sonar and Dataminr. The tools differ, and platform access has changed as companies restricted certain data feeds. The recurring pitch remains stable: turn the public web into an intelligence environment that government can query from one desk.

Platform restrictions do not end the market

Social platforms have occasionally cut vendors off after journalists or civil-liberties groups exposed law-enforcement use. Those interventions matter. They also reveal how much municipal surveillance rests on private access arrangements that residents cannot inspect.

Meta sued Voyager Labs in 2023, alleging that the company used fake accounts and automated collection to gather information from Facebook and Instagram in violation of platform rules. Voyager disputed Meta’s allegations. The litigation concerned the company’s collection methods, not a democratic decision by New Yorkers about what their police department should infer from public speech.

A platform ban can close one pipe while leaving the procurement logic intact. The agency still has trained users, an allocated budget and a demand for broad online visibility. Another vendor may rely on different sources, licensed data or ordinary web collection. The line item changes names.

The institutional appetite survives.

This is why disclosure cannot stop at a city’s current vendor list. Oversight has to follow saved searches, exports, user permissions and downstream records, because those mechanics determine whether a tool supports a narrow investigation or creates standing access to political and social life.

A workable policy would require a documented purpose before a search, restrict monitoring based on protected speech and establish deletion rules that cover exported material as well as the vendor platform. It would also make agencies report the terms they monitor and the number of accounts placed under continuing observation, with safeguards for active cases. Those constraints cost investigators time. That is their purpose.

Friction is one of the few reliable limits on an institution rewarded for collecting more.

The Voyager procurement record is dull by design. It names a vendor, authorizes spending and moves through the city’s machinery. Yet it marks the point where a post written for a loose public becomes available for structured police attention, with relationships inferred around it and the result capable of outliving the feed.

Questions people ask

Can police search public social-media posts without a warrant?

Police can generally view material that anyone can access, although constitutional rules and local policies may constrain particular uses. The harder issue is scale: monitoring software lets agencies aggregate posts, map relationships and preserve results in ways that ordinary viewing does not, sometimes before officers have identified a suspected crime.

What can social-media monitoring software infer about a person?

Depending on the product and available data, it may connect usernames, highlight repeated associations, classify language or suggest relationships among accounts. These outputs are inferences rather than verified facts. Sarcasm, shared hashtags, reposts and casual follows can all acquire investigative weight when presented inside an intelligence interface.

Does deleting a post remove it from a police monitoring system?

Not necessarily. An agency may have saved the post, exported a report, captured a screenshot or transferred the information into another records system. Retention depends on agency policy, vendor settings and local records rules, which is why deleting content from the original platform may not delete the government copy.

How can residents find out whether their city bought this software?

City procurement portals, contract registers, council agendas and public-records requests can identify vendors and spending authority. Search for terms such as social-media analytics, open-source intelligence and threat monitoring, then request policies, user manuals, data-retention schedules and contracts. The bland software line item is often where the surveillance program first becomes visible.