These Are Not Isolated Incidents
Why agentic cyberattacks will keep coming until the industry makes a real investment in defensive and offensive security
· By TEKperfect · 9 min read
On September 21, we closed our AI Alignment series with an appeal for independent standards, independent verification, and real accountability around increasingly automated technology.
We ended on a promising note:
We still have time to build them deliberately.
Eight weeks later, the news has been testing that claim.
A crypto exchange lost more than $6 billion after a fake job interview. A payment network confirmed a breach at one of its suppliers. Supermarket freezers across Europe were switched off remotely. And in a single week, outages hit Microsoft, the FAA, JPMorgan Chase, Apple, and the National Security Agency’s internal communications network.
As the outages piled up this week, one expert called them “isolated incidents.”
We disagree.
They aren’t the work of one attacker, and some of them weren’t attacks at all. But they expose the same weaknesses, and those weaknesses are exactly what AI-driven, agentic attacks are built to find.
What happened in the last six weeks
Binance: a fake recruiter and an internal AI tool
In September, a Binance software engineer was contacted on a job-hunting platform by someone posing as a project director at a non-existent company. After two convincing “interviews,” one with a fake HR agent and one with a fake CEO, the engineer was asked to download a coding assignment.
The assignment contained custom malware. In Binance’s own words, “the 0-day malware was not detected by the endpoint security protection tools at the time.”
By October 20, the attackers had used the engineer’s computer and an internal AI tool to compromise two more employees on the wallet infrastructure team. On October 29, they used those credentials to reach the wallet signing service and moved about $6.34 billion in digital assets across multiple blockchains. Binance says about $2.85 billion has since been frozen or intercepted. The FBI attributed the attack to the North Korean group known as TraderTraitor.
Binance hasn’t said exactly how the AI tool was used. But the lesson is already clear: an AI tool inside your company has access, and anything with access can become part of an attack path.
Visa and JPMorgan Chase: the supplier problem
On November 16, an extortion group called n0n posted 59 organizations to its leak site, including Visa, Mastercard, JPMorgan Chase, Eli Lilly, Walmart, and Procter & Gamble.
Visa confirmed an incident at a third-party provider in Asia Pacific and said its own systems were not involved and that it saw no sign that card numbers were affected. JPMorgan Chase said a service outage the same day was caused by a software defect and that it has no evidence of a cyberattack. As of this writing, n0n’s claims about the bank have not been verified.
What stands out is the pattern. The data in these listings appears to come from vendors that hold it on behalf of the big brands. Expedia confirmed a breach through a third-party vendor just four days earlier.
You can secure your own systems perfectly and still lose your customers’ data through someone else’s.
Europe’s cold chain: old equipment, default passwords
On October 11, attackers reached the refrigeration controllers in as many as 700 French supermarkets and switched them into defrost mode overnight, when stores were empty. No alarms went off. By morning, some cases had reached 50°C. Whole aisles of food were thrown out, some equipment caught fire, and two stores in Alsace alone reported €270,000 in losses.
On November 2, the same kind of attack hit stores in Belgium, Luxembourg, the Netherlands, Sweden, Estonia, Ireland, Portugal, and Germany. Attackers also went after refrigerated plasma at three French blood banks. Staff stepped in after monitoring alerts, and they saved the plasma.
How did they get in? Many of the controllers were reachable from the internet, and many accepted the username “admin” with the current date as the password. The flaws were publicly disclosed in March 2025. Several affected models had reached end of life and would never be patched.
France has blamed Iran’s Revolutionary Guard. But the most important comment came from Richard Horne, head of the UK’s National Cyber Security Centre, who called this “the beginning,” warning that AI can now find and exploit old technology at scale.
A week of outages
Microsoft’s Outlook went down on November 12 because of a bug that had sat undetected since July. On November 13, the FAA grounded flights after a router upgrade overwhelmed its telecommunications network. JPMorgan Chase customers saw delays on November 16. On November 17, Apple services went down after what Apple called “an internal update,” and the NSA’s classified internal communications network suffered what the agency called a “temporary degradation” during scheduled maintenance.
The White House, the Director of National Intelligence, and the companies involved have all said these were not cyberattacks.
We take them at their word. That’s the point.
If a routine upgrade or a months-old bug can take down services that millions of people and our national security apparatus depend on, then a patient, automated adversary looking for the same lever can pull it on purpose.
The common thread
Different attackers. Different industries. Different countries.
The same five weaknesses:
- Trust in people and identities. A convincing recruiter, a stolen credential, or an AI tool with an employee’s access is still the easiest way in.
- Third parties. Visa’s supplier, Expedia’s vendor, and the blood banks’ refrigeration contractor, which blamed “a component supplied by a third-party.”
- Old technology nobody owns. End-of-life controllers with default passwords, still connected to the internet nineteen months after the flaws were made public.
- Concentration. One update, one bug, or one router is able to disrupt millions of people.
- Detection that lags. Malware that endpoint tools missed. Freezers that overheated without triggering an alarm.
None of these weaknesses is new. Security professionals have warned about every one of them for years.
What’s new is who, and what, is now able to exploit them.
Why agentic attacks change the math
In Part 2 of our alignment series, we described AI’s shift from software that responds to software that acts.
That shift applies to attackers too.
An AI agent can scan the internet for exposed refrigeration controllers, try the default credentials, and move on to the next one without getting tired. It can research an engineer’s background and write a tailored recruiter message in seconds. It can sift stolen files from a supplier for the records worth ransoming. And it can do all of this for thousands of targets at once, at almost no cost.
We don’t need to know exactly which of this fall’s incidents involved autonomous tools to see the direction.
The cost of attacking is falling fast.
The cost of defending is not.
And agents now work on both sides of the firewall. The AI tools organizations are deploying to save time have access to email, code, credentials, and internal systems. As Binance learned, those tools can act as stepping stones.
We have under-invested for far too long
For decades, many organizations have treated cybersecurity as a cost to minimize rather than a capability to build.
The results are visible in this fall’s headlines:
- A patch for some of the vulnerable refrigeration controllers was available in August 2025. Many stores never applied it.
- Other models had already reached end of life and were still running critical cold chain equipment.
- Default passwords, a problem the UK has prohibited on consumer-connected devices since 2024, were still protecting commercial equipment across Europe.
- The vendors that hold sensitive data for major brands are often far less protected than the brands themselves.
Underneath the underinvestment is a lack of standards.
There is still no universal, enforceable baseline for the security of commercial connected equipment. The EU’s Cyber Resilience Act won’t fully apply until December 2027. In the United States, security labeling for connected devices is voluntary.
And for AI agents operating inside companies, there is no common standard at all for what they should be allowed to access, how they should be monitored, or how they should be tested.
That is exactly the gap we described in September: without shared standards and independent verification, every organization is left to decide for itself what “secure enough” means. Too many have decided it means “whatever we can afford this quarter.”
What real investment looks like: defense and offense
Defense alone is not enough. Organizations also need to test themselves the way an attacker would.
To be clear, by offensive security we mean authorized testing of your own systems, not “hacking back.”
Defensive investment
- Phishing-resistant two-factor authentication and least-privilege access for every person and every AI tool
- A complete inventory of what’s connected, including building systems, point-of-sale, and refrigeration
- Patching on a schedule, and a budget to replace equipment that can no longer be patched
- Vendor security requirements written into contracts, with the right to verify them
- Monitoring and alerting that works at 2 a.m. on a Sunday
- A tested incident response plan and backups you have actually restored
Offensive investment
- Regular penetration testing of networks, applications, and cloud services
- Social engineering exercises that include fake recruiters and fake vendors
- External attack surface scanning, so you find your own exposed devices before someone else does
- Red teaming of the AI agents you deploy: what can they reach, and what happens if someone else controls them?
Industry-wide investment
- Enforceable minimum security baselines for connected equipment and for the suppliers that handle private data
- Common standards for testing and monitoring AI agents
- Independent verification, so security claims are backed by evidence
- Faster information sharing when one sector is hit, so the next one isn’t surprised
These are the same principles we proposed for frontier AI. They apply just as much to the systems AI is now being used to attack.
Small businesses are the most exposed
It’s easy to read these stories and conclude that only billion-dollar companies are targets.
The opposite is true.
Of the 59 organizations n0n listed on November 16, 19 were small businesses. The supermarkets that lost inventory were individual stores, and two lost €270,000 in a single night.
Small businesses face the same exposures as large enterprises:
- They hire remotely and click on recruiter and vendor messages
- They rely on suppliers and cloud services they can’t inspect
- They run equipment that was installed years ago and never revisited
- They are adopting AI tools quickly, often without anyone deciding what those tools can access
But most have no security team, no one watching alerts overnight, and no budget line for testing. Many are also the “third party” in someone else’s supply chain, which makes them a target for attackers trying to reach their larger customers.
Agentic attacks make this worse. When an AI agent can attack ten thousand small businesses for about the same cost as one, even the smallest organizations can no longer count on being overlooked.
If you run a small business, here is where we would start this month:
- Turn on multi-factor authentication for email, banking, and every admin account.
- List every device and system that connects to the internet, including the ones a contractor installed.
- Change every default password, and replace equipment that no longer receives updates.
- Ask your key vendors how they protect your data, and get the answer in writing.
- Decide what your AI tools can access, and remove everything else.
- Make sure someone gets an alert after hours when something goes wrong.
None of this requires an enterprise budget. It requires deciding that security is part of running the business.
Not isolated. Connected.
A fake job interview. A supplier in another region. A freezer controller with a default password. A software update that took down a national security network.
Looked at one at a time, each can be explained away.
Together, they describe an economy that runs on connected systems we haven’t invested enough to protect; right now, attackers have tools that work at machine speed.
Agentic cyberattacks will keep coming.
The question is whether we keep treating each one as an isolated incident or finally invest in the standards, defense, and offense this moment requires.
At TEKperfect, we believe the time to start was years ago.
The next best time is now.
Sources
- Binance says hackers used fake job interview to compromise computer — Reuters, November 3, 2026
- Binance says fake job interview led to $6.3 billion crypto heist — BleepingComputer
- Visa confirms cyberattack as hackers claim JPMorgan was hit too — Cybernews
- n0n lists Eli Lilly, Visa and JPMorgan: Visa blames supplier — Ransomnews
- 2026 European cold chain hacks — Wikipedia
- Apple outage follows Microsoft, FAA and JPMorgan disruptions — CNBC, November 17, 2026
- Apple, NSA, U.S. banks’ services hit by outages — BNN Bloomberg, November 17, 2026
Not sure where your organization stands?
Take our free 3-minute Cybersecurity Audit, or book a free consultation and we’ll walk through your exposures together.
Book a Free Consultation