An engineer built an AI-guided laser to kill the mosquitoes in his living room. It works — and reveals what we still get wrong.

By David Ogilvy, Chief Marketing Officer at Mosticare Global | Published 2026-06-12

A computer-vision engineer in Cambridge has spent the spring of 2026 turning his living room into a mosquito kill box. Steven Cheng, who describes himself as an expert in computer vision and robotics, posted version 2.0 of his "AI laser mosquito defense system" to X on 28 May. The video shows a rolling chassis the size of a small printer, a DSLR with a long zoom lens, a thermal camera, an aluminium gimbal, and a laser module that pivots, locks on, and zaps mosquitoes out of the air in real time. The whole thing is operated by a deep-learning model trained on annotated mosquito images, with a wide-angle secondary camera that disables the laser the moment a person, a pet, or anything flammable enters the firing arc.

The build is real, the project is genuine, and the engineering is striking. It is also, for the overwhelming majority of households, the wrong answer to a question most of us were not asking. That tension is the actual story.

What Cheng actually built

The system, as described in the Futura-Sciences and Numerama write-ups, has three layers. A high-magnification DSLR scans the room and feeds images to a deep-learning model trained to recognise the silhouette and gait of a mosquito. Once a candidate is identified, a second thermal camera confirms the target, an industrial gimbal aims the laser, and the shot fires. Version 2.0 — the one Cheng announced on 28 May — added harmonic drives, servo motors, and a reinforced aluminium mount to make the aim stable enough to hit a flying insect at speed.

The safety layer is the part the press usually skips. A wide-angle camera watches the room, and a logic gate blocks the laser whenever its firing arc overlaps with a person, an animal, or a flammable surface. Cheng is explicit that the system is not for sale and is not certified for eye safety; the Futura-Sciences piece notes that the closest commercial equivalent, a device called Photonmatrix, advertises up to 30 mosquitoes killed per second within a 6-metre radius but cannot be certified because of the eye-safety risk to humans and animals. Cheng is running his version in his own home, on his own recognisances, with the safety camera as his own line of defence.

The honest summary: this is a working prototype of a technology that has been on drawing boards in university labs for over a decade (Intellectual Ventures' "laser mosquito" concept dates back to the late 2000s), built for the challenge of it, and demonstrated publicly. The headline, "engineer builds AI laser to kill mosquitoes," is true. The implication, "you should buy one," is not.

The wider wave Cheng is part of

What makes the story worth reading is not the device itself but the trend it sits on. Public-health entomology is in a decade of "tech-versus-mosquito" experimentation, and Cheng's build is the citizen-engineer version of a movement that is also running in well-funded laboratories and start-ups.

Sterile-male releases — the same Wolbachia technique covered elsewhere in this brief — are scaling across Europe and the Americas, with the largest proposed US releases now under EPA review. (Italy logged 133 imported dengue cases in early 2026; Valencia's sterile-insect trial has cut local tiger-mosquito populations.) Gene-drive programmes in Tanzania are progressing under African scientific leadership. The XPRIZE-style "vector control" ecosystem, which prizes low-cost, hardware-light, software-heavy solutions, has funded everything from acoustic mosquito sensors to AI species-ID apps. AI is being used in field surveillance to count and identify mosquitoes in trap catches faster than a human ever could.

The pattern is consistent: every new mosquito control idea is being rebuilt around a sensor, a model and a software loop. Cheng's laser is the most photogenic example of the genre, but it is one entry in a much longer list.

What Cheng's prototype actually tells us

Three things, in order of importance.

First, the deep-learning layer works well enough to be the limiting factor, not the enabler. The detection model is described in the French coverage as having "rather good" performance — Cheng has not published a benchmark, but the fact that the system runs in real time and clears a living room in a single night tells you that computer-vision mosquito detection has crossed a practical threshold. That has implications far beyond lasers. Trap counts that used to require a trained entomologist and a microscope are now plausible on a Raspberry Pi with a camera and a small model. Surveillance, the unglamorous plumbing of public health, is the place this technology actually scales.

Second, the killing mechanism is the hard part, and it always will be. A laser is photogenic, but a residential mosquito control that requires safe operation, certification, and consumer-grade reliability is a multi-year regulatory path. The same logic applies to every mosquito-killing hardware idea. The bug zapper never worked; the mosquito racket is a battery fire waiting to happen; the plug-in ultrasonic device is a placebo. Hardware is harder than software in this category, almost without exception.

Third, the user's living room is the wrong unit of intervention. A laser can clear a single room of mosquitoes. It cannot clear the garden, the courtyard, the public park, the school yard, the bus stop. Mosquito control is, at the bottom, a population problem — you need to suppress the Aedes population in the area, not just kill the ones that have wandered into your living room. The Mosticare lens, plainly: the question is not "what kills the mosquito in front of you" but "what stops the next generation of mosquitoes from reaching you at all." Cheng's laser is an answer to the first question, and it is impressive engineering. It is not, and will not be, the answer to the second.

The pragmatic alternative the data keeps pointing to

The combination the field keeps returning to is unglamorous: source reduction (tip out the standing water, screen the gutters, close the saucers under the flowerpots) plus a physical barrier over the place you sleep and the place you sit still. It is the recommendation the French regional health agency gives residents in départements declared "colonised zones" for the tiger mosquito. It is the message behind Italy's ISS season outlook, the Cyprus ECDC "new reality" framing, and the World Health Organization's 2026 insecticide-resistance update. None of it requires a £1,500 gimbal or a deep-learning model, and all of it works whether or not the regulator approves the next experimental control.

None of which is a knock on Cheng. His project is exactly what an engineer with a thermal camera, a long lens, a redundant safety chain and a fondness for the problem should be doing. It is also, and importantly, the kind of project that earns coverage because it is dramatic — and coverage that frames "this is the future of mosquito control" is coverage that quietly under-sells the techniques that already work for almost everyone.

What to watch next

Three signals. First, whether Cheng or anyone else publishes a benchmark for the detection model. The whole "tech-versus-mosquito" pitch rests on a claim of accuracy, and a confusion matrix on a public dataset would move the field. Second, the regulatory status of the Photonmatrix-style commercial laser. The device is sold on Indiegogo from around $498; whether any national safety regulator certifies it will tell us whether the category is a hobbyist niche or a real product class. Third, the deployment numbers on the existing scalable techniques. Wolbachia in Singapore, sterile-male releases in Valencia, gene drives in Tanzania, the Italian arbovirus season — the boring, large-scale, well-run programmes that do most of the public-health work. They do not get AI-gimbal videos, but they are the ones lowering the curve.

A laser that can clear your living room is a wonderful piece of work. A window screen that keeps the tiger mosquito out in the first place is cheaper, safer, and not much less photogenic if you frame it right.

What we know

Sources cited

1. Futura-Sciences — "Moustiques tigres : cet ingénieur a inventé une arme radicale, un canon laser piloté par IA" (Patrick Ruiz, 2 June 2026). https://www.futura-sciences.com/tech/actualites/technologie-moustiques-tigres-cet-ingenieur-invente-arme-radicale-canon-laser-pilote-ia-134959/
2. Numerama — "Il fabrique un canon laser piloté par l'IA pour carboniser les moustiques de son salon." https://www.numerama.com/tech/2264839-il-fabrique-un-canon-laser-pilote-par-lia-pour-carboniser-les-moustiques-de-son-salon.html
3. Les Numériques — "Un ingénieur crée le tueur de moustiques ultime à base de laser et d'IA" (Patrick Ruiz, 2 June 2026, 12:36). https://www.lesnumeriques.com/societe-numerique/un-ingenieur-cree-le-tueur-de-moustiques-ultime-a-base-de-laser-et-d-ia-n256912.html
4. Developpez.com — "Un ingénieur met au point un système de défense laser à base d'IA pour éliminer tous les moustiques d'un domicile." https://intelligence-artificielle.developpez.com/actu/383750/Un-ingenieur-met-au-point-un-systeme-de-defense-laser-a-base-d-IA-pour-eliminer-tous-les-moustiques-d-un-domicile-IA-ingredient-incontournable-pour-deboucher-sur-le-tueur-de-moustiques-ultime/
5. Martin content sweep, 12 June 2026 — item #19. intelligence/martin/2026-06-12-content-sweep.md