decrypted · 18 october 2026 · vulnerabilities and patching · uk policy and law · surveillance and privacy

The Met's unapproved app and the side door in every organisation

For eighteen months, about 500 Metropolitan Police accounts existed on a retail crime platform that the force had never approved. On 14 October the data regulator, the ICO (the Information Commissioner's Office), published a reprimand and an enforcement notice over it. Nobody was hacked. It is the quieter failure that sits behind many hacks: nobody knew which outside services staff were using.

What the ICO found

Between March 2023 and September 2024, officers and staff used Auror, a platform where retailers log shoplifting and share intelligence. The ICO said the force had not formally approved it. About 500 accounts were created and 375 were actively used. Between January and September 2024, personal information on 179 people, three of them under 18, was uploaded to help identify suspected shoplifters.

The ICO concluded the force processed data unlawfully, lacked appropriate technical and organisational measures, completed no data protection impact assessment before high-risk processing, and still lacked measures to ensure only approved third-party technology was used. The notice was issued on 23 September, before the regulator became the Information Commission on 30 September.

How it slipped through

Picture a building with a guarded front door and a visitor book, and a side door anyone can prop open with a sign-up form. The Register reported that the Auror accounts were free, so no purchase ever raised an alarm. It also reported that the force leaned on soft controls such as policy reminders and training, and never tested whether its web controls worked, so the use went unflagged for over eighteen months. Staff themselves raised two internal security reports in August and September 2024, and access was blocked on 26 September 2024.

The Secure by Design lesson

Strip out the badge and this is every UK organisation's problem: software that arrives through a browser, a free trial or a supplier's invitation, outside procurement. The data was ordinary, names and dates of birth and, in a small number of cases, home addresses. That is what an attacker, or a supplier's breach, would want.

The Met accepts the findings. It suspended the platform in September 2024 and brought it back in January 2026 with stronger governance, training, contractual controls and a completed impact assessment. The ICO acted anyway, because the wider approval gap remained.

Approval should be a gate, not a request. That means an approved-services list that is enforced rather than merely published, and a regular test proving the control works. The notice asks for much the same: assess and approve third-party services, review controls, test their effectiveness and report progress. ICO head of investigations David Doodson said "organisations cannot allow innovation to outpace their data protection responsibilities."

The director's question this week: if a team signed up to a new service holding customer or staff data tomorrow, how would we know? If the answer is "someone would tell us", that is a soft control.

Also this week

A Windows flaw is being exploited, and a warning is not a wall. CVE-2026-96951, the public reference number for a flaw fixed in Microsoft's record-sized Patch Tuesday on 13 October, bypasses Mark of the Web, the flag Windows attaches to files from the internet so it can warn you. SecurityWeek reports Microsoft says it is exploited in the wild and that an attacker must convince a user to act. America's cyber agency, CISA, added it to its KEV list, the catalogue of flaws known to be exploited, on 13 October, labelling it a command injection flaw while Microsoft frames it as a security feature bypass. Descriptions differ; the advice does not: patch, and do not rely on users noticing a missing prompt.

The data regulator turns to AI agents. On 8 October the ICO said ten foundation model developers, including Amazon, Anthropic, Apple, Google, Meta, Microsoft and OpenAI, had made or committed to changes on transparency, people's rights and safeguard assessments. It also opened a six-week call for evidence on agentic AI, closing on 20 November 2026, and said it is questioning OpenAI, Anthropic, Meta and the AI Security Institute about agent testing and deployment. Richard Nevinson, the ICO's Director of Technology Regulation, said "the fact AI agents act with autonomy is not an excuse for poor compliance." If you are giving agents access to personal data, you have until 20 November to be heard.

H&M says there is no sign of compromise, but the claims point at a supplier. The group ShinyHunters, and later one calling itself ShinyEquationCartel, claimed a breach of H&M and leaked a small sample. BankInfoSecurity reports the sample appears tied to GreyOrange robotic fulfilment systems used in H&M's North American warehouses. H&M said "we have not found any indications that our data environment has been compromised." The claims are unverified. If accurate, the lesson matches the lead: your data estate includes every tool your suppliers run on your behalf, so ask who holds your data and how you would learn of a problem.

Sources