When AIs go to war: Autonomous cyber weapons ‘inevitable’

Cyber attacks perpetrated by artificial intelligences (AIs) operating autonomously from human oversight are basically inevitable, and security professionals urgently need to start thinking about how to adapt their policies to deal with this, according to Trend Micro’s vice-president of security research, Rik Ferguson.

Speaking at the supplier’s annual CloudSec event in London, Ferguson said that today, AI in the enterprise has not really advanced far beyond machine learning techniques for analysing large data pools, and is certainly nowhere near reaching Hal 9000 or Skynet-like capabilities.

However, given the virtual necessity of using some form of machine learning to manage legitimate data pools, it would be foolish to think that malicious actors are not also using it to exploit illicitly acquired personal or corporate data, said Ferguson.

“AI is like any other tool – something that can be used for good, and something that can be repurposed for different ends,” he said.

“AI doesn’t think like a human. We chain ourselves to our own preconceptions, but AI is free to think in totally different ways. When threat-modelling from a cyber security perspective, you are suddenly going to find yourself in a world where you have to take AI models of thinking into account.”

Ferguson told his audience that once AI can be leveraged properly by cyber criminals, CISOs can expect their organisations to be subject to AI attacks that move much faster, contain context-aware malicious code, and adapt themselves to impersonate different people within an organisation to attract the attention of a target in a phishing scam, or pretend to be a standard penetration testing tool to gain access to a network.

Ferguson gave the example of the widespread and dangerous Emotet trojan, which, since it emerged five years ago, has become, to some extent, able to adapt itself to its circumstances – lying dormant if it finds itself accidentally in a sandbox, or behaving differently when running in a virtual machine environment in an attempt to cover its tracks.

“Emotet is not AI-driven right now, but imagine if it was,” said Ferguson. “Imagine how rapidly you could adapt your threats and attacks. For you to be facing an autonomous cyber weapon is pretty much inevitable.

“It’s time we need to be thinking in a world of AI versus AI – how do we deal with that threat?”

Theresa Payton, president and CEO of security consultancy Fortalice, who was the first ever woman to hold the post of White House CIO under president George W Bush, said that although more traditional cyber crime models were still very effective, the advent of genuine AI posed serious questions, and the industry as a whole was falling behind.

“My concern is that the Mirai botnet was just a test,” said Payton. “Once we get to 5G and autonomous everything and AI-led cyber crime scenarios, where you could potentially have AI launching its own attacks without human intervention – where do you go from there? What are the strategies?”

Robert McCardle, director of Trend Micro’s Forward Looking Threat Research team, said that responding to these new attacks would be a challenge because cyber security and law enforcement agencies were both handicapped to some extent because they have to “move at the speed of the law”.

For example, making requests for co-operation with foreign agencies requires specific processes and channels, something cyber criminals are not subject to.

“If you want to see an exercise in how the as-a-service model works, look to cyber criminals,” said McCardle. “They’ve got trust models, escrow, everything. And they move faster than the defenders. The defenders are handcuffed and the people who should be wearing the handcuffs have no problem.”