In March 2026, Reuters reported that the Pentagon used Anthropic’s AI services, including Claude, during US strikes on Iran. That’s the same AI assistant people use to draft emails, review documents and write code. Reuters cited a source familiar with the operation, but couldn’t establish what Claude actually did.
When AI becomes part of a weapons system
A test in Ukraine in late September gave us a clearer example. Reuters reported that AI software developed by the Dutch company Intelic identified an incoming projectile as a threat and helped coordinate the response. A person made the final decision to launch an interceptor drone.
When we talk about AI being used as a weapon, it’s worth looking at what happens before anyone gives the order to fire. In the Ukrainian test, the software had already made an assessment that could influence the person authorising the launch. If that assessment is wrong, how much time does the person have to spot the mistake? And what information do they have to make up their own mind?
What AI changes for hackers
In its September 2026 threat report, Anthropic described an espionage campaign it linked to Russia, alongside cases involving financially motivated attackers. The company said it observed people using Claude to carry out parts of their operations and process stolen information. People still chose whom to target. Anthropic says it disrupted the activity it identified.
For a small business, the worrying part is how many organisations a single attacker can go after. It’s easy to think nobody would spend much time trying to break into a business they’ve never heard of. But when some of the work can be automated, a business can get caught up in a campaign simply because it uses the same software as hundreds of others.
Why UK government websites went offline
On 22 October, The Independent reported that 32 UK government services were temporarily unavailable and another four were partly unavailable. The website of the Office for Budget Responsibility was among those affected. The disruption followed the discovery of serious flaws in PHP, a programming language used by websites around the world. A security researcher had used AI to help uncover the problem.
The government later told Parliament that it had taken some public-facing services offline as a precaution while teams applied updates and other protections. It said there had been no reported compromises of government websites as a result of the vulnerability. Finding the flaw gave the teams responsible for those services a chance to act before anyone was known to have exploited it on a government site.
Fake public support and fake personal connections
In its report on influence operations, Anthropic described how an operator in Kenya used Claude to prepare social media posts ahead of the country’s 2027 election. The posts were meant to look as though they came from ordinary people expressing their own views. Think about the last time you read the comments under a political story. If several people seem to agree, it’s easy to assume the view is widely held. Anthropic found that much of the influence content it investigated drew little or no genuine engagement.
The same report describes a network of more than 20 dating apps run by a studio based in China. Over two weeks in April, more than 4,700 AI personas spoke with at least 25,000 people. The apps were promoted as places to meet real people, and some profiles did belong to real people. It’s one thing to be suspicious of a message from a stranger. It’s harder when you’ve been chatting for days and the person on the other end remembers what you told them yesterday.
The €95 million impersonation scam
In September, Reuters reported that fraudsters used AI to impersonate senior executives and steal €95 million from Fideuram, the private banking arm of Italy’s Intesa Sanpaolo. According to two sources familiar with the case, the person who received the request believed it was genuine and instructed the finance department to arrange a series of transfers.
By the time the instruction reached the finance team, it was coming from someone who genuinely believed it was legitimate. That’s what makes this case worth paying attention to. A colleague can pass on a fraudulent request in good faith. If an unusual payment instruction comes from someone senior, confirm it through your usual approval process or call them on a number you already have. It’s easier for staff to make that check when everyone knows it’s part of the job, even when the request comes from the boss.
Could AI have helped in the supermarket freezer attacks?
In October, attackers remotely interfered with refrigeration controls at more than 700 retail sites in France. The French government attributed the sabotage to an Iranian cyber unit and said it caused fires and spoiled food worth tens of millions of euros. Reuters reported a further wave of incidents across Europe in early November. In Ireland, wholesaler Musgrave said 13 SuperValu stores were affected and some chilled and frozen products had to be withdrawn from sale.

As Euronews explored in its coverage of the freezer attacks, AI could help attackers find weaknesses in systems like these. Whether it helped in this case remains speculation, as does exactly how the attackers gained remote access. For the people running those shops, the consequence was immediate. They had food they could no longer sell and customers due through the doors.
The $6.3 billion Binance hack
In late October 2026, hackers stole about $6.3 billion in digital assets from Binance, the world’s largest cryptocurrency exchange. The FBI attributed the theft to North Korea, identifying a state-sponsored group known as TraderTraitor. Although it might be described as a Bitcoin hack, the target was the exchange. Reuters reported that Binance had used its own funds to buy about $6 billion in replacement tokens. FBI
The case also shows how the cost of a cyberattack can grow after the theft itself. Binance had to spend billions replacing assets that customers had entrusted to the exchange. The FBI’s public attribution identifies who it believes carried out the theft. Whether AI helped the group remains unknown
At AI Tuition Hub, we offer a number of courses to help people recognise scams, protect personal information and reduce the risk of fraud. We also examine how AI is being used to fight crime, from spotting unusual transactions to helping investigators review evidence, and where people still need to check the facts before acting.
Published: 13th November 2026.