Claude AI: Why are there so many internet outages?

This week, AI chatbot Claude went down, leaving users unable to access the service via its maker Anthropic’s website, but barely a week goes by without a similar incident at a technology giant, government website or hospital. What’s causing this apparent uptick in problems?

One of the main vulnerabilities of the modern internet is the shift to cloud computing, meaning a huge range of websites and services now rely on just a handful of companies, such as Amazon and Microsoft. In the early days of the commercial internet in the 1990s, companies used to operate their own hardware and software, a bit like individual shops in a street. If one of those companies had a problem, their shop would close, but the rest would be unaffected.

These days, companies are far more likely to host all their operations on the cloud, which is like the street’s road, sewer system and electrical grid rolled into one. If that goes down, then all of the shops are out of action and we all hear about it.

Sometimes, these problems can be caused by simple human error. Nothing highlights the danger of this sort of incident better than the 2024 outage when cybersecurity firm CrowdStrike released a software configuration file that took down millions of Windows computers worldwide, knocking airlines, banks, television companies and emergency-service call centres offline.

Joseph Jarnecki at the Royal United Services Institute, a UK defence think tank, says that if an outage is large and its effects wide-ranging, it is unlikely to be deliberate. Ransomware criminals, who break into systems and lock up data before demanding payment, know better than to tussle with huge technology companies full of experts – they go after smaller prey.

Tim Stevens at King’s College London says that ransomware attacks are increasingly targeting small local governments and infrastructure. Their business model is to break in, lock up something that people rely on and demand a ransom, so what better to target than a town’s water supply, electricity grid or local government?

In the UK, we have seen exactly that, with ransomware attacks against Hackney Council, Gloucester City Council and Leicester City Council, as well as the NHS and water suppliers. Stevens says that for as long as we have had computers, there has been a cat-and-mouse game between hackers and security experts. Unfortunately, at the moment, the hackers are ahead. “I’ve heard, in the last year or so, more people than normal from the profession saying that we’re losing. Not just that we’re behind, but we’re actually losing.”

State-backed hackers from countries like Russia and China are also unlikely to take down an entire cloud provider. “They definitely target them, but not to destroy and disrupt,” says Jarnecki. “They’re incredibly highly targeted.”

An example of this would be the 2023 attack on Microsoft-run US government email accounts, which were hacked by what Microsoft said was a China-linked group. The wider service was largely unaffected, but spies got access to a treasure trove of US secrets.

Sarah Kreps at Cornell University in New York says that targeted cyberattacks are also used by nations in what is now called the grey zone – a state of tension that isn’t quite peace and isn’t quite war, but is a carefully considered and measured tussle that stops just short of causing all-out conflict.

“This is a form of economic sanction in a way, because so much of our GDP, our economic welfare, relies on the internet. If you can take that down, you’re handicapping the adversaries’ ability to generate wealth. And the ability to generate wealth is how you develop the resources to fund a war, to fund allies in a war,” she says.

Kreps points out that Russia and China aren’t the only ones doing this. While we occasionally hear about Western cyber warfare – GCHQ and MI6 famously hacked into computers belonging to al-Qaeda and changed a recipe for bombs into one for cupcakes – it is happening regularly, but is highly classified and done behind closed doors.

“My understanding, based on interactions with the US intelligence community, is that that is going on,” says Kreps. “You do have an incentive to erode the strength of an adversary. There’s a good motive behind [attacks on] Russia for their involvement in Ukraine and there’s a good motive for trying to erode China’s capabilities as they become a peer competitor.”

Stevens says that Western countries are constrained in the scale and target of their cyberattacks because, unlike some nations, they are bound by a strong rule of law. “I have no doubt whatsoever that our intelligence agencies and our security services in general are conducting operations in cyber against Russian assets,” says Stevens. “But it’s hard work and there are lawyers always in the room and we are somewhat constrained. I think there’s a lot of frustration about that.”

While Claude is back up and running now, Anthropic didn’t respond to questions from New Scientist on the cause of the outage.