I can see it now: the Terminator travels back to 2021 and then casually walks by the office for Boston Dynamics, Tesla, even X1, and Figure AI, and instead stops in front of Anthropic. With his characteristic Austrian accent, the Terminator flexes his formidable muscles and says, “I must stop programmers from makingz Claude AI. I vill prevent the first almost fully automated, large-scale cyberattack in 2025.” Meanwhile, the robot developers scurry away, figuring the Terminator might be back in a decade for their hides.
In real life, there is no Terminator, but there are extremely worrying signs about AI’s rapid development and new concerns about its weaponization. This week, Anthropic revealed that it mostly thwarted a massive “AI-orchestrated cyber espionage campaign.”
- Are now more intelligent
- Have agency where they can take autonomous actions, chain them together, and even make decisions with little human input
- Can even use tools on your behalf to search the web and retrieve data
Hackers using AI to turbocharge their efforts is not new. Even the spam texts and phone calls you receive every day are accelerating because AI makes it easier to spin out new IDs and strategies.
However, these more recent advancements appear to be helping hackers attack at scale and with little more than some very basic programming and, primarily, prompts.
According to Anthropic, “Overall, the threat actor was able to use AI to perform 80-90% of the campaign, with human intervention required only sporadically.”
The only good news is that Anthropic detected the activity and quickly shut it down before it got far enough to have any noticeable real-world impact.
The next, scary level
Still, these hackers were highly motivated and quite canny. They got around Anthropic’s safeguards by breaking the attack down into tiny, innocuous pieces, tasks that separately seemed harmless, but together composed the full attack.
As I see it, everything Anthropic shared about this cyberattack is deeply concerning and should be read as a warning for all of us.
The rapid pace of AI development means that these platforms will only get smarter. Agentic AI, in particular – models that can carry out tasks on your behalf – are on the leading edge of development of virtually all AI platforms, including those from Google (see SIMA 2, which can play in virtual worlds on its own) and OpenAI, and while most of it is used for good, these capabilities are clearly tantalizing for cyber attackers.
It might seem like this is purely a concern for governments, businesses, and infrastructure, but the breakdown of any of these systems and companies can quite often lead to loss of services, support, resources, and protections for consumers.
So, yes, Skynet is still the fictional big bad of our AI nightmares, but it won’t take a robot army to bring down society, just a hacker or two with access to the best AI has to offer. The next Terminator will surely be visiting AI companies first.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
Add Comment