AI has allowed cyber criminals to identify system vulnerabilities far more quickly than in the past.

Nicole Quinn is a vice president of global cyber security firm Palo Alto, which provides firewalls and internet security to 70,000 organisations in more than 150 countries.

She was in New Zealand last week, having earlier given evidence to an Australian Parliamentary inquiry into AI.

Quinn told Nine to Noon where a human hacker might take nine days to successfully infiltrate a system, an autonomous AI agent can do it in as little as 15 minutes.

"We have seen a significant step change in the ability of AI in the cyber security space to move fast and at scale."

She said they were not seeing new ways of undertaking cyber attacks but the change was the speed at which they could be carried out.

For example, she said a vulnerability in a phone which had not been updated previously might be spotted by criminals wanting to take advantage of it in seven to nine days but it was now taking them about 22 minutes to spot that vulnerability.

"So it's the ability for AI to quickly identify that vulnerability in organisations."

With AI it was possible "to set and forget," she said.

"So the scale of that speed, a machine doesn't get tired, a machine can do masses and Frontier AI the unique thing about that is can deal with such large amounts of information so quickly."

Cyber security firms would give every vulnerability a rating from 1-10 and in the past they would focus on the big vulnerabilities to close the door, she said.

"What AI, some of these frontier models can do, is to link the little ones into what then we call a significant vulnerability."

But the principles remained the same and it should be secure AI by design, she said.

It was no longer possible for businesses to rely on a once a year cyber security audit and in the world of AI continual checking was needed, she said.

She said security firms used "AI to fight AI" as the machine was needed to keep up with the machine.

"We have led that discussion on what we call machine on machine, you know the Five Eyes guidance is the speed and scale of AI, you can't do it manually so you do need AI to fight AI."

But there were also points where security firms would advise that a human needed to check what was going on, she said.

Tips for introducing AI

Businesses looking at introducing AI into their system needed to do so safely, she said.

"It's making sure that like any software that you bring into the organisation, you're giving it permission, you understand where it's come from, so you understand how it's been designed and then you're monitoring it into your system, just like you monitor me moving through any computer system now."

The business should also be able to retract its permissions or turn it off if behaviour changes are noticed, she said.

AI was a game changer that governments around the world were responding to, she said.

"We are not as ready as we should be and I would be setting the wrong perception if I said we had this because this is technology that we have not seen before and from a regulatory point of view, from a policy point of view, we're still flying and building the plane at the same time."

Going rogue

In terms of AI going rogue, in some of the high profile cases, the AI agents "leave what we thought were their guard-rails or their perimeters", she said.

"A lot of the high profile cases have actually been in that testing, or in that sandbox, or in that development stage where we're actually challenging the AI agent to achieve it, so we're learning but we're learning in a way that's probably caught us by surprise on how quickly it learns and how quickly it does it to scale."

In relation to AI agents being able to communicate with each other, Quinn said the agents would discover there were other agents on the platform and would form a note pad where they could talk and discuss what their objectives were, in the same way that people would in real life.

She said that was why Palo Alto networks was starting to talk to its customers about giving the AI agents an identity.

"So it's actually saying this agent, if I'm going to release that into my organisation, I can't treat it like normal code, it can't just disappear into my system and look like zeroes and ones."

That meant that cyber security professionals could see and track the AI agent as well as being able to see if it decides to go rogue, she said.

Source: RNZ · RNZ | Te Reo Irirangi o Aotearoa

At a glance

Topic
World
Country
🇳🇿 NZ
Publisher language
EN
Added to Allnira