by Tom Evslin
The problem will often be AI used with the best of intentions and for good purposes. Advanced AI models are uncovering serious vulnerabilities in existing systems and even in the tools that were used to create these systems. These vulnerabilities have been around for decades in some cases but remained undiscovered or too difficult to exploit without the help of AI.
Now that the owners of these systems know about these flaws and know that hackers using AI tools are likely to discover them soon or learn about them on the dark web, there is a mad dash to fix the problems and put the fixes into production. AI can create patched code at lightning speed; but each patch has the potential to create new problems that can cause systems to crash even without a push from hackers.
We programmers have always spent more time testing and debugging than actually writing code. AI can write code many times faster than we humans ever could; it can generate test cases and even run them with great efficiency. It can look for logic errors in code. What AI doesn’t do well (at least not yet) is anticipate strange cases where an innocent user error, a hardware malfunction, a communication glitch, or some combination of these can cause a system to spiral into a malfunctioning morass of attempting to correct errors while actually propagating error conditions or shut down in an autoimmune reaction to what it perceives as an attack.
So we get lots of new code fast, compelling reasons to put the new code into production immediately before the system gets hacked, and we don’t take enough time to roll it out node-by-node and wait for weird bugs that not even AI has thought of to surface. Even when there appears to be a problem, we don’t quickly fall back to the old system, which we now know is known to be vulnerable to hacking.
We’re going to see this a lot in the weeks and months to come. Sometimes, hopefully not usually, the error will be the side effect of a fix. Sometimes there actually will be a hack. There will certainly be hackers spamming you with urgent requests for your passwords and other information claiming to fix a real or non-existent problem.
Be careful. Whether in a plane (assuming they’re flying) or not, buckle your seatbelt. Turbulence has been reported ahead. Good luck to us all in these interesting times!
Retired hitech entrepreneur. Active AI experimenter (github.com/tevslin). Developer of SmartTranscripts

