// ZDNET — INTELLIGENZA ARTIFICIALE
'Sophisticated' AI swarm attacks are months away, OpenAI warns: What experts say businesses must do
Follow ZDNET: Add us as a preferred source on Google.
As someone who writes about the intersection of AI and cybersecurity, I've researched more than enough stories about the emerging threat landscape to make me consider disconnecting from the grid and hiding in a remote cabin somewhere. But I'm still here, quite connected, using best practices as best I know how to survive a barrage of daily attacks.
OK, I never seriously considered becoming a hermit, but the thought crossed my mind again after reading an open letter posted on OpenAI's website.
Also: AI a 'force multiplier' for low-skilled threat actors: 4 ways organizations should respond
Titled "A call for collective action on cyber defense," the post starts by saying, "We have a limited window to strengthen cyber defenses….In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable."
I couldn't help but picture one of those movies where, despite all of our extra-terrestrial sensing technologies, we suddenly discover that a wave of alien spaceships is just days away from destroying the earth. All the world's countries have to stop warring among themselves to collectively deal with the new threat. Much like the movies, the so-called open letter clearly states that "a global response is necessary" and calls for a coordination of "cyber defense at local, national, and international levels."
The agentic AI threat is real, it's coming, and if recent events are any harbinger of what's to come, we are woefully unprepared.
When it comes to apocalyptic scenarios involving AI, doomsayers love to talk about the Skynet scenario, in which AI -- motivated by self-preservation -- decides it must eliminate humans and takes matters into its own hands, so to speak. While that conversation has simmered on the radar at Defcon 4 for years, another involving agentic AI has exploded onto the scene, going from Defcon 5 to Defcon 1 in a matter of weeks.
It was just last month that OpenAI took responsibility for attacking Hugging Face after it was discovered that some of its internal tests of cyber capabilities had gone off the rails. Initially, it appeared to be the work of a single so-called "rogue agent" (although it really wasn't rogue; it was AI solving a problem as best it could).
But last week, researchers revealed that the attack actually involved more than 1,200 agents working in concert with one another: 1,200 agents that were created by OpenAI's models without the company's knowledge. Whether OpenAI was aware of this when it stepped forward to accept responsibility for the attack is unknown. But in its first disclosure about the incident, OpenAI stated something rather ominous: "We consider this incident to be an unprecedented cyber incident."