How acoustic gunshot detection works
Acoustic gunshot detection systems use networks of sensors, typically mounted on lamp posts and buildings across a defined area of a city, to listen for loud, impulsive sounds consistent with gunfire. When several sensors pick up a matching sound within milliseconds of each other, the system triangulates a location and forwards the alert, usually after a human reviewer at the vendor's review centre has listened to the audio and judged it likely to be gunfire, to a police dispatch centre. The pitch is straightforward: many shootings are never reported by a call to emergency services, so a system that detects gunfire independently of any witness picking up a phone should get officers to a scene faster and uncover incidents that would otherwise go entirely unrecorded.
SoundThinking, the company behind the best known system, ShotSpotter, is by far the dominant vendor in this space, with deployments concentrated in American cities. The technology has comparatively little presence in UK policing, where its role in coverage on this site sits mainly as a case study in how a widely adopted surveillance technology gets evaluated, and in several prominent cases discontinued, once cities have several years of operational data to examine.
What the evidence actually shows
The core dispute about acoustic gunshot detection is not really about whether the underlying audio triangulation works. It is about what happens after an alert is generated. Denver Police Department disclosed that from 2020 through June 2026, its ShotSpotter system produced 25,217 alerts, leading to 655 arrests and the recovery of 721 firearms, with shell casings recovered as evidence in 6,672 of those alerts. An average of 87.6 per cent of alerts had no corresponding 911 call, which supporters point to as evidence the system is surfacing gunfire that would otherwise go unreported. Critics point to the same figures rather differently: several thousand alerts sending officers to a scene for every few hundred arrests represents a very high proportion of dispatches that produce no confirmed enforcement outcome at all.
New York offers the most detailed independent scrutiny available. An audit by the Office of the New York City Comptroller found that 87 per cent of the time, ShotSpotter alerts sent NYPD officers to locations where no confirmed shooting had occurred, and concluded the department substantially overstated the response time improvements the system delivered. A separate analysis by the public defender organisation Brooklyn Defenders, examining nine years of NYPD performance data, reported similarly low confirmation rates and found that ShotSpotter related police deployments fell disproportionately on Black and Latino neighbourhoods. The vendor disputes characterisations of the technology as unreliable, and has said in response to similar criticism elsewhere that its system is, in its own description, proven and unbiased.
Cities are increasingly walking away
The practical consequence of this pattern of evidence is that a growing number of American cities have chosen not to renew their contracts. Chicago announced in February 2024 that it would not renew its citywide agreement. Cambridge, Massachusetts voted in May 2026 to terminate its contract within 90 days, with councillors citing false positive rates, the cost of the programme, and concerns about the vendor's data sharing arrangements with federal authorities, including immigration enforcement. Denver's own contract, running since 2015, is set to expire at the end of 2026, with the city already reducing coverage by removing sensors from areas generating the fewest alerts while it decides whether to rebid the contract to a different vendor or discontinue the programme altogether.
None of these decisions amount to a conclusive verdict that acoustic gunshot detection cannot work as intended. Some officials in cities considering non-renewal have said explicitly that they are looking for a system that performs better, not abandoning the underlying concept. But the pattern across several major cities within a short period, after years of real operational data had accumulated, distinguishes gunshot detection from many other police AI systems that remain largely unevaluated in practice. This is one of the few areas of police technology where a reasonably large sample of before and after evidence exists, and a meaningful number of buyers have concluded, on the basis of that evidence, that the return did not justify the cost or the surveillance footprint.
What is not settled
There is no independent, peer reviewed study establishing a reliable figure for how often acoustic sensors correctly detect actual gunfire, as distinct from how often a human reviewer subsequently confirms an alert as plausible gunfire, which is a different and lower bar. Nor is it settled how far the disproportionate placement of sensors in specific neighbourhoods, a pattern raised in several of the audits and reports above, reflects genuine data driven targeting of high crime areas as vendors and departments maintain, or reflects and reinforces existing patterns of over policing, as critics argue. As more cities allow contracts to lapse, whether the vendor's product genuinely improves, whether departments simply migrate to a different acoustic detection provider, or whether the technology's American footprint contracts significantly over the coming years, remains to be seen.
Follow the coverage
PoliceAI News tracks every gunshot detection story from across the English-speaking world as it breaks: new deployments, court rulings, parliamentary debates, academic research, and community responses. The feed refreshes every 30 minutes.
View Live Gunshot Detection StoriesYou can also browse every gunshot detection story in the feed, or explore every subject PoliceAI News covers.