BadBIOS: sounds awfully spooky. Worse than Cylons. I estimate the odds of an indestructible multi-platform virus that can jump air gaps being a real thing at less than 1%, but I wasn’t overly Bayesian in my estimate. See also the relevant discussion on Reddit. What would be your estimate of such a spooky thing being real?
I think you misinterpreted the article. The virus can’t infect a healthy machine “through the air” (microphone). It can bridge air gaps in the sense that two already infected machines can setup network over microphone, which is orders of magnitude more likely than the former. BIOS infections have been done before, so …
I estimate the odds of an indestructible multi-platform virus that can jump air gaps being a real thing at less than 1%
I estimate the odds that there’s a virus, which has all the capabilities BadBIOS is claimed to have (acoustic networking, spread through USB stick firmware, regeneration from system firmware), at 99.9%. All of the pieces have published precedent, and there are multiple published non-secret meta-malware packages for putting pieces like that together.
Conditional on malware like that existing, I’d estimate the probability that Dragos Ruiu having malware like that at 80% (the main alternative is that he’s got malware, but he’s wrong about its communication channels or mechanism of spread). However, at this point conditional on him having it, I’d estimate the probability of him successfully procuring a complete sample (as opposed to a loader module) at only 50%.
I think the main alternative is that he has paranoid delusions and no malware at all. It seems unlikely that he would incorrectly conclude that a piece of ordinary malware has all these advanced capabilities.
BadBIOS: sounds awfully spooky. Worse than Cylons. I estimate the odds of an indestructible multi-platform virus that can jump air gaps being a real thing at less than 1%, but I wasn’t overly Bayesian in my estimate. See also the relevant discussion on Reddit. What would be your estimate of such a spooky thing being real?
I think you misinterpreted the article. The virus can’t infect a healthy machine “through the air” (microphone). It can bridge air gaps in the sense that two already infected machines can setup network over microphone, which is orders of magnitude more likely than the former. BIOS infections have been done before, so …
Yeah, they discuss this point. Still, even acoustic healing sounds rather hard, though not impossible.
Hearing about this makes me fear the unboxability of AI even more
I estimate the odds that there’s a virus, which has all the capabilities BadBIOS is claimed to have (acoustic networking, spread through USB stick firmware, regeneration from system firmware), at 99.9%. All of the pieces have published precedent, and there are multiple published non-secret meta-malware packages for putting pieces like that together.
Conditional on malware like that existing, I’d estimate the probability that Dragos Ruiu having malware like that at 80% (the main alternative is that he’s got malware, but he’s wrong about its communication channels or mechanism of spread). However, at this point conditional on him having it, I’d estimate the probability of him successfully procuring a complete sample (as opposed to a loader module) at only 50%.
Gwern created a predictionbook entry for this question.
I think the main alternative is that he has paranoid delusions and no malware at all. It seems unlikely that he would incorrectly conclude that a piece of ordinary malware has all these advanced capabilities.
He being wrong about what kind of malware he has doesn’t mean he has ordinary malware.