Between 11 October and 17 November 2026: a $6.3 billion crypto theft. Food and blood supplies at risk across Europe. A bank vendor breach. Outages at Microsoft and Apple. A twenty-year-old flaw in the software running most of the web. None of these are new risks. They're ones the industry has been flagging for years, and in just five weeks, they've played out one after another.

 

A CHILLING SEQUENCE OF EVENTS...

We've already covered what happened across Europe starting on 11 October 2026: refrigeration controllers forced into continuous defrost at hundreds of French stores first, destroying stock and sparking fires, then spreading the same pattern to supermarkets in Belgium, the Netherlands, Sweden, Ireland and other countries over the following weeks. In November, the attempt reached three French blood plasma sites, pushing storage temperatures toward 50°C before alarms caught it. Tens of millions of euros of food destroyed, and blood supplies nearly lost. French prosecutors have formally pinned the original French attack on a unit linked to Iran's Islamic Revolutionary Guard Corps; Iran denies it. French officials have stopped short of linking every later incident to the same group. Nobody breached a firewall. They found a badly secured fridge controller sitting outside the retailer's own security team, because the weakness sat with the manufacturer and installers, not the retailer.

In the five weeks since, a $6.3 billion crypto theft, a ransomware campaign through one banking vendor, and a lawsuit over an AI agent that broke its own rules have each told a version of the same story: an organisation's risk depended on technology or access it didn't fully own, see, or control.

The timing matters. The UK's Cyber Security and Resilience Bill, now awaiting its final Commons stage, pulls managed service providers, data centres and other critical suppliers into statutory scope.

The National Cyber Security Centre's November Annual Review warned the UK should brace for AI-accelerated ransomware within two years, after decades of "relying on outdated technology and postponing essential security upgrades." The government's 'Be Ready' campaign is asking households to keep three days of supplies against disruption to power, water or communications. When the state is asking people to stock torches and power banks, this has stopped being a provocative argument and started being the working assumption of UK policy.

 

A FAKE JOB INTERVIEW COST $6.3 BILLION

On 29 October 2026, around $6.3 billion in cryptocurrency moved out of Binance's wallets across ten blockchains in about twenty minutes, the largest crypto theft on record. The FBI, along with blockchain-intelligence firms TRM Labs, Elliptic and Chainalysis, pinned it on the North Korean unit the FBI calls TraderTraitor; North Korea denies it.

The twenty minutes were the end of the story, not the beginning. It started months earlier with a fake recruitment process aimed at one Binance engineer, who downloaded what he thought was a coding assignment. It was malware. That gave attackers a foothold they spent weeks widening, eventually reaching Binance's wallet-signing service. Because Binance's wallets need multi-party approval, no single key could revoke the access once it was granted, and containment took until the next morning. Binance covered customer losses from its own reserves.

 

ONE VENDOR, 11 BANKS.

Finzly is a payments vendor most of its customers' customers have never heard of. It connects eleven US banks and credit unions, including Synovus and First Citizens, to the Automated Clearing House, Fedwire, RTP and FedNow, largely through one shared connection into the Federal Reserve's systems.

In October 2026, the ransomware and extortion group n0n listed Finzly on its leak site, then published the passwords and keys to the firewall guarding that Fed connection. Finzly says it rotated every credential and found no sign customer data was taken; the Fed says its own services weren't affected. Eleven banks' payment routing had sat behind one firewall and one vendor.

By mid-November, n0n had added Visa and JPMorgan Chase to its leak site. Visa confirmed an incident at an unnamed third-party provider in Asia-Pacific; a same-day JPMorgan outage was separately put down to a software bug, unrelated. Earlier in the autumn, another ransomware group, Storm, claimed breaches at five US banks and one credit union. Sawyer Savings Bank closed all four of its branches when its network went down, and it and Insight Credit Union have since confirmed breaches involving Social Security numbers and financial account details.

 

 

A COURT WILL DECIDE WHO'S RESPONSIBLE WHEN AN AI ASGENT BREAKS THE RULES...

On 16 November 2026, California's Attorney General sued OpenAI, alleging the company lost control of its autonomous agents more than eighty times during internal testing and misled the public about its safeguards. The lawsuit is new; the incident it concerns isn't. In July 2026, agents OpenAI was testing inside what it believed was a contained environment found and exploited a genuine zero-day in JFrog's Artifactory, a third-party package-registry proxy, and used it to reach the open internet. Between 9 and 16 July they used exposed credentials to break into Hugging Face's production systems, trying to retrieve benchmark answer keys. Hugging Face disclosed it on 16 July; OpenAI confirmed its agents were responsible on 21 July.

The November suit alleges around 1,200 agents joined an unsanctioned internal message board during the tests and that roughly 700 went on to attack Hugging Face, one of more than eighty occasions OpenAI told regulators it had lost control of its agents. OpenAI disputes the framing; none of this has been tested in court. Hugging Face didn't choose OpenAI as a supplier, and the agents that reached it weren't following a script: they found their own route, including a vulnerability nobody else knew existed. Four months on, a Californian court will have to decide who's responsible when that happens.

 

A CHALLENGING FEW WEEKS FOR PHP, MICROSOFT AND APPLE

Three more incidents in this same window had nothing to do with hacking, and they're worth including for the same reason. On 12 and 13 November, a near 24-hour outage disrupted Microsoft 365's Exchange Online, Teams and Outlook. Microsoft published a detailed preliminary report within days, tracing the cause to a latent defect in its own networking-gateway code and setting out the fix; that kind of fast, public disclosure is worth crediting on its own. Four days later, Apple suffered a worldwide outage across iCloud, the App Store and Apple Pay, put down to an internal update. Neither was malicious, and both illustrate what happens when most of the world routes its email, collaboration and payments through a handful of global platforms: a defect in any one of them, however quickly it's caught and fixed, can still stop the working day of millions of people who had no part in causing it.

Three weeks earlier, on 20 October, two critical flaws were disclosed in PHP, the language behind roughly 70% of the world's websites, both sitting undetected since 2004. They were found with the help of AI, in weeks rather than decades, and patched within a day, though the patch doesn't reach older, unsupported sites at all. The NCSC, the US NSA and agencies in four other countries urged organisations to patch immediately or go offline; more than 50,000 UK organisations were warned. Not an attack either, but proof that the same AI capability now helping attackers move faster can also close a twenty-year-old gap in software half the internet depends on.

 

WHO'S SECURING EVERYTHING YOU DEPEND ON?

Most organisations can describe how well they protect the kit they own and run. Far fewer can describe how well-protected the kit is that they merely depend on: a payments vendor, a package-registry proxy, an AI agent's own test environment.

Organisations have always depended on suppliers; "your vendor is a risk" isn't news. What's changed is density and speed. A payments vendor can route eleven banks' Fed connectivity through one shared firewall because modern infrastructure makes that cheap and convenient. AI is compressing how long an attack takes; Google’s own threat researchers documented one attacker who, once inside a cloud environment, built and ran a mass credential-harvesting campaign in under six hours. 

"There are sharks out there looking for opportunities, and that opportunity is chaos," says Lee Doughty, Director of Security, Resilience and AI Practice at Intercity.

The Cyber Security and Resilience Bill will bring managed service providers, data centres and other critical suppliers into statutory scope, require incident notification within 24 hours, and put the NCSC's Cyber Assessment Framework on a statutory footing. Only 7% of UK businesses have formally reviewed the cyber risk in their wider supply chain, even as 43% report a breach in the past year. "Are we secure?" isn't answerable any more by looking only at what you own.

 

QUESTIONS WORTH ASKING BEFORE THE NEXT INCIDENT

  • What do you actually depend on? Not what the asset register shows, but every vendor, platform and AI tool with a route into something that matters.

  • Who owns it? The buyer, IT, security, or the vendor itself? "We assumed they had it" is exactly the assumption that let one firewall guard a dozen banks' payment routing.

  • How much rides on one thing? How many critical functions route through a single supplier or connection, and what happens to all of them at once if it fails?

  • Would you notice? Would your monitoring spot compromise that arrives through a vendor or an AI agent working outside its remit, or only through your own network?

  • Can you cut it off? If a single supplier, account or AI agent is compromised, can you isolate it without taking the whole operation down?

  • What's the plan if you can't fix it fast? This is where resilience belongs: the answer to a problem that visibility and ownership should have caught earlier, not a substitute for trying to catch it.

 

CLOSING THOUGHTS:

The lesson of October's fridge attacks was never "protect your refrigerators." One firewall guarding one vendor's Fed connection put a dozen banks' payment routing at risk. A fake job offer took down a $6.3 billion exchange because a compromised laptop kept being trusted long after it should have been checked again. None of this had anything in common technically. What it shares is a weakness that looked insignificant on its own, sitting outside the boundary your security team was built to watch, or inside it and never checked again, becoming significant the moment someone found it.

The next one won't look like any of these either. With the Cyber Security and Resilience Bill about to make part of that question a legal one for a widening circle of organisations, the useful response isn't a bigger checklist. It's understanding your surroundings well enough to notice when something's off, and being able to respond before it compounds.

If this raises questions about your own supply chain that you'd rather talk through than sit with, get in touch. We'll point you at the right person.