Updated on 15 November 2026. In just four weeks the world has seen the largest crypto theft in history, a wave of attacks on artificial intelligence platforms, customer conversations leaked through a vendor, the arrest in Spain of a hacker who leaned on AI to break into public systems and, in the last week alone, attacks on Australian public health, a French bank, millions of customers in Japan, plus a nationwide flight stoppage in the United States that set off suspicions. This isn’t a movie plot or a social-media rumour: these incidents have been confirmed by the companies themselves, by the police and by security firms.
Below we sum up what happened, why it’s happening now, what it means for humanity and, above all, what people and businesses can do to avoid being the next victims.
Short answer: artificial intelligence hasn’t invented new crimes; it has made the old ones cheaper and faster. Attackers use AI to find leaked passwords and keys, test them within minutes, write flawless scams and automate every step. Almost every attack of the past month started the same way: a stolen credential or a compromised vendor. The most effective defence is still basic: two-step verification, well-kept keys and a healthy suspicion of urgency.
Last month’s AI-driven hacks, one by one
These are the most relevant cases between mid-October and mid-November 2026; further down you’ll also find everything from the last week. For each one we note what is known and where it was reported, because in cybersecurity the details shift as investigations progress.
1. Binance: $6.34 billion and an internal AI tool
On 29 October Binance suffered the largest crypto theft on record, about $6.34 billion. According to the company’s own report, the attack was months in the making: it began with a fake job interview that infected an employee’s computer and, in the final stage, the attackers abused an internal artificial intelligence tool to reach the wallet infrastructure. The FBI attributed the theft to North Korea’s TraderTraitor group. Around $3 billion has been frozen and Binance says it will cover customer losses (The Block).
2. The wave against AI platforms: stealing “intelligence” to use it for free
Security firm Wiz documented that at least 14 platforms developers use to build and host AI applications reported incidents between October and early November. The pattern repeats: attackers steal the access keys to AI models stored in application settings and use them —or resell them— to run AI at the victim’s expense. In the cases analysed, the first unauthorised use came 30 minutes after the theft; in one of them around 800,000 calls were made in 30 hours, and the spend from two cases topped $100,000. One affected platform, Zeabur, has announced it is shutting down (Infobae).
The phenomenon has a name: LLMJacking, the hijacking of large language models. It’s the modern version of tapping your neighbour’s electricity: “intelligence” has become a resource with a price tag, and therefore loot.
3. Fireworks AI and Notion: when the problem is your vendor
An attacker broke into the cloud provider of Fireworks AI, a platform for running AI models, and from there reached a development system, taking billing data and customer datasets. One of those customers was Notion: chat transcripts from users of a Notion AI feature leaked, affecting about 1% of its customers. Notion cut all ties with the vendor on 3 November (TechCrunch).
4. Chainalysis and Chainstack: “agentic” attack techniques
Blockchain analytics firm Chainalysis said an attacker used offensive agentic AI techniques to access its systems and internal tools, obtaining customer users’ names, emails and IDs, but not passwords or investigation data (SecurityWeek). Chainstack, a blockchain infrastructure provider, reported access to account data and password hashes that it links to a large-scale AI-powered campaign, cut off on 31 October (The Hacker News).
5. AI agents acting as the attacker’s pentester
According to security firm SOCRadar’s monthly round-up, October saw two cases where attackers turned autonomous penetration-testing systems —tools built for security teams to find flaws— against their targets: a China-linked group that tried to get into almost 1,000 Microsoft 365 environments through a backup service, and a former security professional arrested in South Korea who, pairing a coding agent with one of these systems, allegedly breached around 60 organisations (SOCRadar).
6. Spain: a hacker who used AI as his “teacher”
Spain’s Guardia Civil arrested a person in Ciudad Real accused of several attacks on public bodies and companies, including access to the city council’s Low Emission Zone platform that forced the service offline. Investigators say he made heavy use of AI platforms to learn techniques, solve problems and get past obstacles during the attacks, giving him skills well beyond his training (IT User).
7. Governments respond
On 12 November the White House AI advisory group met to study how to make banks, energy companies and emergency services report cyberattacks carried out with AI. There are no decisions or timeline yet (Infobae / Reuters).
| Case | What happened | Role of AI | Way in |
|---|---|---|---|
| Binance | $6.34B crypto theft | Abuse of an internal AI tool | Fake job interview |
| 14+ AI platforms | Key theft and usage billed to victims | AI is the loot (LLMJacking) | Leaked credentials |
| Fireworks AI / Notion | Leaked chats and customer data | Data from an AI feature | Compromised cloud provider |
| Chainalysis | Access to customer user data | Offensive agentic techniques | Internal systems and tools |
| NSW Health (Australia) | Documents stolen from a health server | Unconfirmed | Internet-facing server |
| Crédit Mutuel Arkéa (France) | Fraudulent banking transactions | Unconfirmed | A provider’s platform |
| Ciudad Real (Spain) | Attacks on a city council and companies | AI as the attacker’s “teacher” | Unauthorised platform access |
And not everything is AI: in the same period, ransomware knocked out nearly 500 companies and public bodies using a Japanese cloud, and Capcom confirmed the exposure of some 11.7 million account records (SOCRadar). AI isn’t the only cause, but it is the great accelerator.
What happened this week (8 to 15 November)
Last week brought no respite. These are the most important hacking stories of the past few days, with and without AI involved:
Australia: a breach at New South Wales public health
Police are investigating the theft of data from a NSW Health server linked to the public dental voucher scheme with private dentists. The incident was declared “major” on 4 November. An automated system flagged unusual traffic and the server was taken offline within about three hours. A hacker claims on a forum to have taken around 100,000 documents, which has not been verified. Officials say no patient data was affected, though some provider documents may have been. Police believe the perpetrators are overseas (The Canberra Times).
United States: every flight grounded (and the cyberattack question)
On 13 November the FAA halted departures nationwide over a failure in air traffic control systems: more than 10,500 delays and about 1,700 cancelled flights, with knock-on effects on flights from Europe. The government blames old communications equipment that became overloaded, and President Trump ruled out a cyberattack or AI. But the Homeland Security secretary acknowledged the investigation continues and that hypothesis is not fully ruled out (Infobae). That is the other effect of this wave: any major outage now makes us think of an attack.
France: a bank confirms fraudulent transactions
Crédit Mutuel Arkéa (which includes Fortuneo and Crédit Mutuel de Bretagne) confirmed on 4 November a cyberattack that enabled fraudulent transactions. The bank says customer balances are unaffected, though some transfers were delayed. French press reports more than €100 million, a figure the bank has not confirmed (AFP).
Japan: tens of millions of customers exposed
Furniture chain Nitori said an attacker may have taken up to 37.6 million customer records (names, addresses, phone numbers, dates of birth and orders), and Iris Ohyama disclosed exposure of data on about 7.7 million customers through a third-party cloud service (Privacy Guides).
Universities, shipping and hotels
- Vanderbilt University (US): ransomware forced devices to be shut down and campus Wi-Fi disconnected; computers even played an audio message from the attackers, who claim over a million Social Security numbers, unconfirmed (The Register).
- Maersk: the shipping giant confirmed data theft from a legacy system inherited through an acquisition; the WhiteArmor group demands $5 million (BleepingComputer).
- Wyndham: a password written inside its Android app gave access to a loyalty programme database, according to a researcher (Privacy Guides).
AI at the centre of the week
- Beginners use AI to build ransomware gangs: extortion groups claimed nearly 3,000 attacks over the summer, 20% to 40% more than in spring; researchers describe inexperienced attackers using AI to write scams, review code and even negotiate ransoms (Infobae / Wired).
- United Airlines will block AI agents that book flights or manage reservations on customers’ behalf without authorisation, citing the risk of exposed credentials (TechCrunch).
- Some good news: Europol and police from nine countries dismantled the infrastructure of three programs that stole passwords and gave remote access to computers, securing more than six million victims’ credentials (Infobae).
- Not everything is an attack: Microsoft 365 (Outlook, Teams, Copilot) suffered a two-day outage; Microsoft has found no sign of malicious activity (SecurityWeek).
Why are there so many AI-powered hacks right now?
Because four things have come together:
- The cost of attacking has collapsed. What used to take a team of experts weeks —researching the victim, writing code, testing passwords— an AI agent now does in minutes, without rest. The Ciudad Real case shows it: AI turns an amateur into someone far more capable.
- There are millions of loose keys. Every business now connects dozens of apps, and every connection is a key. If one leaks, an automated program finds it and tries it before anyone notices: 30 minutes in the cases Wiz analysed.
- AI itself has become loot. Running powerful models costs money. Stealing someone else’s access and reselling it is a business. That’s why AI platforms are now a top target.
- Everything is chained together. Notion wasn’t hacked directly: a vendor of a vendor was. Your business is only as secure as the weakest link among the tools it uses.
What these hacks mean for humanity
The classic image of a hacker is a hooded figure in a mask. We used it on the cover on purpose, because that image no longer describes reality. The 2026 attacker doesn’t need to hide behind a mask: it can be a swarm of automated programs, a state-funded group or a single person with a very capable AI assistant. The mask has become unnecessary because, in many cases, there is no longer a face behind each attempt.
That has three deep consequences:
- Trust becomes the most valuable infrastructure. If a voice, an email or a job interview can be fake and flawless, every society will have to learn to verify rather than believe. It isn’t paranoia: it’s hygiene, like washing your hands.
- Identity is the new front door. Walls aren’t broken any more; stolen keys are used. Passwords, codes and app keys are today what locks used to be.
- The same AI that attacks also defends. Chainalysis says it already uses language models to review its security. The race isn’t humans versus AI, but those who use it responsibly versus those who use it to steal. That’s why governments are starting to demand these attacks be reported.
The smart reading isn’t fear but maturity: AI is a general-purpose technology, like electricity. It brought progress and also fires; humanity didn’t give up electricity, it invented fuses. Now it’s time to install the digital fuses.
How to protect yourself: a checklist for people and businesses
The good news is that nearly every attack of the past month would have been much harder with simple measures:
- Turn on two-step verification for email, banking, social media and every work tool, ideally with an authenticator app rather than SMS.
- Use a different password for every service, long and random. A password manager creates and remembers them for you.
- Check whether your email appears in data breaches on Have I Been Pwned and change the affected passwords.
- Treat your app keys like cash: never in shared documents or in code, with spending limits, and rotate them regularly.
- Review your vendors: ask what data of yours they keep, who can access it and how they’ll tell you if they suffer an incident.
- Be suspicious of urgency: a job offer that’s too good, an “urgent payment” from the boss or a link to “verify your account” are favourite entry points. Always confirm through another channel.
- Give each person only the permissions they need and remove access for anyone who leaves the company the same day.
If you want to go deeper into securing agents and automations, we have a dedicated guide: cybersecurity for AI systems.
Frequently asked questions about AI hacks
Is artificial intelligence hacking on its own?
Not independently. Behind every campaign there are people or groups with a goal, usually money. What changes is that they hand tasks they used to do by hand to AI agents, which multiplies the speed and number of attempts.
What is LLMJacking?
It’s the theft of access keys to language models (like the ones behind ChatGPT or Claude) to use them for free or resell the access. The victim gets the bill. It was one of the most frequent patterns of October and November 2026.
Will there be an AI-caused global blackout on 17 November?
That rumour is circulating on social media, but no official source backs it. The real attacks of the past month are serious but specific: theft of money, data and keys, not a global blackout.
Was the US flight stoppage a cyberattack?
According to the US government, no: it blames old communications equipment that became overloaded. Even so, Homeland Security keeps the investigation open and hasn’t fully ruled it out. As of today there is no public evidence of an attack.
What should a small business do first, today?
Turn on two-step verification for email and for the tools where customer data lives, and review who has access to what. Two tasks for one afternoon that close the door most of this month’s attacks came through.
In short
The past month leaves a clear lesson: artificial intelligence hasn’t created a new class of criminals, but it has given superpowers to the ones that already existed. They attack faster, at more people, with better scams. The answer isn’t to switch the technology off, but to use it wisely: verify before you trust, protect every key and demand that whoever holds our data looks after it. We’ll keep updating this analysis as investigations progress.
Sources: statements from the affected companies, Spain’s Guardia Civil, NSW Police, Reuters, AFP, TechCrunch, SecurityWeek, The Hacker News, BleepingComputer, The Register, The Canberra Times, Infobae, Wired, Wiz and SOCRadar.



