decrypted · 26 july 2026 · ransomware and cybercrime · vulnerabilities and patching · uk policy and law

The Craneware breach, and the 2,000 hospitals waiting on Edinburgh

Craneware, the Edinburgh-headquartered, AIM-listed maker of billing and pricing software used by more than 2,000 US hospitals and around 10,000 clinics and pharmacies, told the London Stock Exchange on Monday that hackers had accessed "a subset of its data environment." They copied a significant volume of file names, some employee data, and a slice of customer and partner records. The company has notified the Information Commissioner's Office and the FBI, engaged external forensics, and says operations were never disrupted. Its shares fell nearly 10% anyway. This is a useful story precisely because it isn't dramatic: Craneware appears to have done most of what good incident response looks like, and still ended up here.

What was actually taken

Craneware is careful to say a lot of what was copied is "non-sensitive or already public" data, plus file names rather than confirmed file contents. Don't read that as nothing happened. An attacker who walks away with an index of file names across a shared billing platform has effectively photographed the labels on every folder in the cabinet, even without opening one. For a company that sits between thousands of hospitals and their revenue, knowing which folders exist, and for which named customers, is itself useful reconnaissance for a follow-on phishing or business email compromise run against any of those 2,000 hospitals, dressed up as routine Craneware correspondence.

The design lesson, not the disclosure lesson

The fact that this stayed a "subset" of the environment is evidence that some segmentation was already in place, and that's worth crediting rather than assuming failure. But when the product is one shared platform serving thousands of downstream healthcare organisations, a "subset" is still a large blast radius measured against any single hospital's own risk appetite. The Secure by Design principle at stake here is tenant isolation, not just functional isolation: a vendor holding data for many customers should be architected so that a compromise of its own corporate environment cannot cross into any individual customer's data by default, not by policy. Quick disclosure to the ICO and FBI, and bringing in outside forensics fast, is genuinely good practice. But that's the response bar. The design bar is making sure the tenant boundary held before any of that becomes necessary. UK software firms selling into regulated sectors overseas, healthcare especially, should treat this as the standard to build to, not just the standard to be judged against afterwards.

Also this week

A critical Check Point zero-day, and the internet-facing management console problem. CVE-2026-16232, an authentication bypass in Check Point's Security Management and Multi-Domain Management products, lets an unauthenticated attacker obtain a login token with full administrator rights via SmartConsole. CISA added it to its Known Exploited Vulnerabilities catalogue on 22 July, rating it 9.1, with a 25 July deadline for federal agencies. Check Point says it has only affected "a handful of customers," all sharing one configuration: a Management Server exposed directly to the internet without IP restrictions. Patches are out. The sharper lesson is that this bug only bites organisations that made a management console reachable from the open internet in the first place, exactly the kind of choice Secure by Design is meant to make the hard default, not the easy one.

Two Scattered Spider members jailed over the Transport for London hack. Thalha Jubair, 20, and Owen Flowers, 18, were sentenced to five years and six months each at Woolwich Crown Court on 16 July for the 2024 attack that cost TfL an estimated £29 million and forced 27,000 staff to reset passwords in person. The National Crime Agency called it the largest cybercrime prosecution ever brought before UK courts. They didn't break any cryptography: they rang TfL's IT helpdesk with partially stolen credentials and talked their way past it. Any UK organisation whose helpdesk can reset a password or wave through an MFA re-enrolment over the phone should treat this verdict as the prompt to check whether that process is actually harder to social-engineer than it was in 2024, not just better documented.

Four AI agent security disclosures in July, one shared flaw. Separately published research this month covered a hijackable Claude for Chrome extension, a single email able to implant false long-term memory in an agent, an open-weight model backdoored with ten tainted training examples for under £75 (demonstrated by UK researcher Katie Paxton-Fear), and AI connector ecosystems that reconfigure every few minutes. None of these are model safety failures. All of them are trust-boundary failures: what an agent will click, remember, inherit or connect to without question. That's the frame UK organisations piloting agentic AI should already be applying, well before the procurement questionnaire asks about model alignment.

Sources

If you're building software that handles other people's data and want a second opinion on where your trust boundaries actually sit, get in touch.

More like this

Get the next post by email: subscribe to Decrypted. Double opt-in, unsubscribe any time, or take the RSS feed.