Dubai Police announced 36 robots. None of them have bodies.

On 6 September 2026, Dubai Police announced that it had deployed 36 robots over the past year. The story reached this site three times before the day was out, under three headlines. One of them used the word robocops.

The force's own announcement is more specific. The 36 are robotic process automation robots, deployed across administrative and operational functions. Robotic process automation is software. It has no body, no sensors and no presence on any street.

What was actually announced

The occasion was an annual inspection of Dubai Police's General Department of Artificial Intelligence by the force's Commander-in-Chief. The figures presented were these: 104 technology projects completed over the year, 36 RPA robots deployed, public service continuity of 99.97 per cent, system continuity of 99.92 per cent, seven awards, and 29 registered intellectual works.

Also reviewed were an AI platform supporting staff and the public, and a smart assistant used in the force's Smart Police Stations.

Read as a whole, this is an IT directorate's annual report. Uptime figures, project counts, an automation programme, an internal assistant. It is a reasonable set of things for a large police organisation to be doing, and reporting uptime at all is more disclosure than many forces manage.

What it is not is a story about machines patrolling anywhere.

What robotic process automation actually does

RPA is software that performs repetitive clerical work by driving other software. It copies a value from one system into another. It updates a record when a form arrives. It reconciles two databases that were never designed to talk to each other, by doing exactly what a human clerk would do, at the same interface, faster and without pausing.

The technology has been sold under the word "robot" since the mid-2010s, long before the current wave of AI. It is one of the least glamorous products in enterprise software, and one of the most widely bought, because most large organisations run on systems that cannot exchange data directly.

The word is a marketing inheritance. It does not describe a machine.

Why the word does so much work in policing

In most industries, "we deployed 36 RPA robots" is understood immediately as an automation project and produces no coverage at all.

In policing it collides with an image that already exists. Everyone knows what a police robot looks like, because a film established the picture in 1987. So a sentence that would be inert in a banking press release becomes a headline about robocops, and an announcement about back-office automation is read as an announcement about the street.

Nobody needs to have lied for this to happen. The vocabulary does the work on its own.

The 2017 promise, and the 2026 number

The comparison worth drawing is with Dubai Police's own history, because this is the force that produced the original image.

In May 2017 it unveiled what it described as the world's first operational robot police officer: a customised REEM unit built by the Spanish firm PAL Robotics, roughly 1.7 metres tall, moving on wheels, with a touchscreen on its chest for reporting crimes and paying fines, and facial recognition capable of flagging matches to a control room. The stated aim was for robots to make up a quarter of the force by 2030. Officials noted at the time that such machines take no "sick leave or maternity leave".

Other announcements followed the same year, including an autonomous patrol vehicle with a companion drone, and a much larger rideable robot promised for Expo 2020.

Four years short of that 2030 target, the robot figure the force chose to publicise is a count of software automation scripts.

That is worth stating carefully. It does not establish that the physical robot programme was abandoned, and no such conclusion follows from a single announcement. Dubai Police runs Smart Police Stations and an extensive camera network, and this site's tracker carries its Oyoon system, which combines facial and plate recognition across the city. The physical machines may still exist, and simply not have been what this year's inspection covered.

But a headline number is a choice about what an organisation considers its strongest evidence of progress. When the flagship robot figure moves from humanoids on the street to automation scripts in the back office, that shift is itself informative, whatever else remains in service.

The categorisation problem this creates

For anyone maintaining a record of police AI, the interesting question is not whether a headline oversold something. It is where this belongs.

This site filed all three versions of the story under Robotics and Autonomous Systems. On the substance, that is the wrong shelf. Nothing in the announcement relates to police ground robots. Administrative automation belongs alongside case management and back-office tooling, and the automated part of it raises questions about auditability and error handling, not about machines in public space.

The aggregation followed the word rather than the substance, which is precisely the failure the piece above describes, occurring on this site. Saying so is more useful than quietly recategorising it.

It also matters for counting. Three archive entries for one announcement makes a routine departmental review look like a developing story. Volume of coverage is not weight of evidence, and any record built from feeds has to keep those two apart.

The general case

This is the third distinct way a police AI claim can mislead without anyone being dishonest, and they are worth naming together.

• A system can be procured, designed and funded without being deployed, so contracts get read as capabilities.
• A match can be reported as an identification, so a lead gets read as a fact.
• And a product can carry a name that describes something else entirely, so software gets read as a machine.

All three produce the same result: a public picture of police AI that is more advanced, more physical and more certain than the evidence supports. Correcting them does not require scepticism about the technology. It requires reading the primary source.

What would be worth reporting

There is real material in Dubai worth establishing, and none of it is the robot count.

How many Smart Police Stations operate, and what proportion of public contact runs through them. What the smart assistant does when it cannot answer, and whether a person is reachable. What Oyoon's watchlist consists of, on what legal basis, and with what retention. Whether any accuracy or error figures for any of it are published.

Those questions concern a system that demonstrably affects people in Dubai. The 36 robots do not, unless you happen to work in the department whose paperwork they now handle.

Which raises a question worth asking of every police AI announcement, including the ones that sound impressive: if you removed the word that made this a story, would there still be a story?