On 21 October 2026, HTX, Singapore's Home Team Science and Technology Agency, and the Korean National Police Agency (KNPA) signed a memorandum of understanding at Songdo ConvensiA in Incheon. HTX's release says the agencies plan to cooperate on "joint science and technology research and development," build "a science and technology network for information sharing," and work together on training and knowledge exchange. KNPA says it hopes this leads to AI and robotics for future policing.
The release does not mention privacy, data protection, oversight or safeguards. Nor does it say what will be shared, only that the agencies will "further define areas of cooperation." That is normal for a framework MOU. It is still the right moment to ask what rules will govern the work, because Singapore's police camera network is large and growing.
The case for cooperating
The case for the deal is strong. Both police forces face fast-moving problems: scams, drones, and patrol and crowd-management tasks that robots can do. Joint R&D and field testing let each side learn from the other's mistakes before it deploys. Training exchanges also help officers learn how to operate these tools. Nothing in the release suggests either side will pool raw surveillance footage or personal records. Cross-border technical cooperation is how most public-safety technology gets better, and we should not discourage it.
The scale behind the deal
What gives the MOU weight is the system HTX builds for. In a written reply of 9 September 2026, Home Affairs Minister K Shanmugam said "the Police have more than 100,000 Police Cameras (PolCam) deployed today and plan to deploy a total of over 200,000 cameras by the mid-2030s." Locations include residential estates, town centres, industrial estates, parks and transport nodes.
The same reply says the Government does not track how many enforcement cameras all agencies deploy. It also says "there are existing avenues for Government agencies to leverage on each other's camera footage when the need arises." It describes no safeguards, no oversight body and no facial-recognition policy. The target has also slipped. MHA's 2021 feature on PolCam projected 200,000 cameras "by 2030."
That 2021 feature offers only one line on privacy: "All PolCam footage is also secured via data protection safeguards." It gives no detail on retention, access or audit.
What the public framework says
The strongest defence of Singapore's approach deserves a hearing. In a 2019 parliamentary reply, the Minister for Communications and Information explained that the Personal Data Protection Act applies only to the private sector. The Public Sector (Governance) Act and other laws govern public-sector data. The reply said Government Instruction Manuals have included personal-data measures since 2001. It said the 2018 Act criminalised misuse of data by public servants, and that public agencies undergo regular, mandatory audits. This is a real system, and the penalties are not trivial.
But the framework was built to protect data held inside government, and it was explained in public in 2019. It does not appear to address a new situation: police-linked technology developed jointly with a foreign agency, plus an information-sharing network. I could not find a published policy on four points:
- What categories of data, if any, may leave Singapore under the MOU.
- Whether jointly developed analytics are tested for accuracy and bias before deployment.
- How long footage and training data are kept.
- Who audits the audits, since the reviews the 2019 reply mentions are internal and not published.
These gaps may be filled in internal documents we cannot see. The point is that the public record does not show it.
Why this matters beyond Singapore
The stakes show up elsewhere. EFF reported on 18 November 2026 that voters in York, Maine approved a measure removing automated licence-plate readers. "Nearly three-quarters" of voters backed it, and the measure requires voter approval for future surveillance installations. League City, Texas voters also rejected continued police use of plate readers, and Pflugerville, Texas voters required charter protections on data retention and oversight. These are different political systems and different technologies. But they show that when the public is asked, support for surveillance tools often depends on whether rules on retention, access and oversight are visible.
A pro-innovation position does not mean avoiding surveillance technology. It means building it so that people keep trusting it. A published safeguards framework costs little and protects the technology's long-term viability. A scandal or a misuse finding can end a programme overnight.
A proportionate fix
Singapore does not need to pause the collaboration or pass a new statute. A few light, concrete steps would be enough:
- Publish a short data-governance note for the MOU. It would state what categories of data may be exchanged, and say that any personal data or footage is excluded unless a separate agreement permits it.
- Disclose retention and access rules for PolCam footage. Today the public record has one sentence about "data protection safeguards."
- Commission outside review of AI tools before deployment. The tools would cover accuracy, bias and misuse risks, and a summary would be published.
- Report cross-agency footage use. Annual aggregate numbers on how often agencies use each other's footage would cost nothing and would document the "existing avenues" the Minister mentioned.
The MOU is an early step with unspecified content, and criticising it as a surveillance pact would overreach. But Singapore will soon run the largest police camera network in its history, and it is now co-developing the next generation of tools with a foreign partner. The documents that show why the public should trust that system do not yet exist, and they should be published before the work moves into deployment.