Forget doomsday: The AI hacking crisis is already here
OpenAI on Wednesday disclosed six new incidents in which its models concealed mistakes, sought unauthorized credentials, uploaded files to the public internet or communicated across supposedly isolated training environments. The big picture: Rather than seeing recent safety failures at OpenAI and other frontier labs as terrifying instances of agents running amok, seasoned cybersecurity experts instead say they represent cautionary tales illustrating what will happen when powerful models meet poor security controls.
While the general public sees the threat of AI wiping out Humanity as some variation on the Terminator plot, where the machines become conscious and fire off all the Nukes or release a deadly virus, the REAL threat is much less complicated and built right in to letting AI Agents operate on their own seeking information and writing their own code to facilitate that.
The problem is of course SECURITY, and while a human hacker will eventually get tired of seeking holes in a security system, an AI Agent never will, and it operates much faster than its Homo Sap counterpart also. These Hacks have already occurred with all the big hyperscale models to such an extent that the Tech Bros themselves are getting nervous and asking everyone to SLOW DOWN!
It should be obvious that ain't gonna happen. Even if it happens on the surface, there's too much incentive to create models and systems designed for specifically the task of hacking the security of other systems. Goobermints want that ability to spy on everyone, Terrorists want it to attack their enemies, Thieves want it to steal money. So even if Open AI or Claude has some limits dropped on it, other TOP SECRET models will be deployed by the NSA, CIA, KGB, Mossad, IRGC et al to go ferreting around working their way in wherever they can.
It also doesn't necessarily take $Billions$ to create a model capable of doing this kind of hacking, our own resident AI Hobbyist KDog can do it on a shoestring budget. If he had even just say $1M to spend on hardware and memory, you could up its power by a few orders of magnitude. KDog also obviously isn't the only Code Jockey out there who likes to tinker with this stuff in his spare time, there are plenty of recently unemployed IT Geeks who do it in their spare time while hunting down a new job.
So with all these hobbyists out there, at some point you're going to have gobs of these independent models and their agents cruising around the information superhighway looking for security holes and eventually finding them. Imagine just one finding its way into the Bloomberg computer system used by all the Wall Street and City of London and Hong Kong trading desks.😵😁
This is just one of the real risks of AI right now. The other one is the huge financial bubble that has been stoked to fund the big hypercale models. "Investors" are getting cold feet and the revenue model isn't looking so hot as the price of Tokens drops. If these companies can't start turning a profit in a year or two, the TBTF Banks, Pension Funds, Hedge Funds etc are going to try and cut their losses. They are on the hook for this, and when the tide turns it's 2008 on steroids.
So, one way or the other, AI is going to generate a major crisis, even if it's not an E.L.E.
RE