From our other sites
The story
Sparked by comments from OpenAI’s Sam Altman that “fixing every bug is no longer possible,” a discussion about AI and security spread across 5ch’s Science News+ board. With the GMO ransomware extortion case and the panic over attacks exploiting PHP vulnerabilities fresh in everyone’s mind, the thread’s real focus was the underlying structural problem: the speed at which AI finds vulnerabilities has outpaced the speed at which they can be patched. Centered on the asymmetry where attackers only need to exploit a single hole while defenders must plug every single one, opinions diverged on why banks and crypto exchanges suffer such different scales of damage, and on how Japan’s position — long shielded by its language barrier — is now changing.
Sam Altman: “There’s no way we can fix every single bug anymore”
Watching the GMO ransomware extortion case and the panic over attacks targeting PHP vulnerabilities, you might think Japan is at the center of all the cyberattacks right now — but that’s far from the whole picture.
Look abroad and you’ll find case after case where AI is uncovering security holes (vulnerabilities) massive enough to break the internet on a daily basis — and before those holes can even be patched, AI-automated operations are already breaking in through them!
Source: gizmodo.jp / Original article here
What people said
He was already a billionaire before OpenAI
Uber
Airbnb
Stripe
Guzzles electricity and memory like crazy
It's exactly because it *can't* be done that we're in this mess
People always say defenders are at a disadvantage
Of course finding just one hole to attack is faster than finding and fixing tens of thousands of them
Back in the day, they lined cannons along the coast to guard against the Black Ships.
That's when Sakamoto Ryoma's mentor said:
"We should just get Black Ships of our own." (the Black Ships were the foreign warships that forced Japan open in the 1800s — i.e., fight fire with fire)
Sam Altman's basically just an investor to begin with lol
But what if the AIs team up with each other instead?
They'll realize how pointless fighting is
Get the reward function wrong and that's exactly what happens
Even with humans, when money-worship takes over you get cases where the censoring body colludes with what it's supposed to be censoring
The same thing happens with AI if you mess up the reward settings for the side doing the checking
If our power gets taken away, humanity's activity gets taken away with it
Once you steal crypto you can do whatever you want with it freely, but with banks you'd also have to rewrite the accounting system itself — breaking in and transferring money out isn't easy
That's why the only hits are things like individual passwords getting stolen at the edges
It's not that it suddenly became impossible — it always was
If there were a vulnerability that let you make transfers, though, seems like huge amounts of money would get stolen in an instant
That might be true, but first-time transfer destinations, large amounts, and overseas transfers are actually still approved by human hands, so that's another hurdle you'd have to clear — not so simple
With crypto, once it's stolen and the blockchain's rewritten, it's over, but with banks there are still hurdles before the money can actually be used, so if it gets caught partway through, the attack fails
No way
Human approval is such a sieve that it still lets all those scam-call frauds through (tokushu sagi — Japan's notorious phone/wire fraud scams, often targeting the elderly)
Whoaaaa, analog really is the strongest, huh
There are even cases where removing the bug makes it stop working entirely
Analog's the best
In France, a supermarket's freezers got hacked and all the food had to be thrown out
that's gotta be the biggest story out there
Because the profit outweighs the damage
with massive personal-data leaks
I doubt it tops Binance's trillion-yen heist
Go look it up yourself
Opening a bank account these days is an absolute pain
Especially once you're over 50
Mizuho grilled me like crazy
And all I wanted was to open an account because I'd just moved nearby
Most of these 'unauthorized access' cases might just be AI browsing data through improper methods because someone asked it to 'look into this in more detail'
Most companies are only saying 'unauthorized access = there's a possibility information leaked,' nothing more
And the AI being used for that unauthorized access is the one Sam Altman is supplying
Even once a hole is found, it takes time to build a patch, so the defender is at an overwhelming disadvantage.
It's like a sterile, zero-immunity environment suddenly getting hit by a virus — and the result is total chaos, screams of agony all around.
Background and Key Points of This Discussion
In the context of cyberattacks, it’s long been pointed out that there’s a basic asymmetry: attackers win by finding just one hole, while defenders lose unless they plug every single one. The idea underlying this discussion is that AI has widened that gap further by speeding up vulnerability discovery. Thread opinions split over why banks have escaped the kind of critical damage crypto exchanges have suffered — some argued it’s less about system robustness itself and more about the slow, human-approval process required for high-value or overseas transfers acting as a brake, while others pushed back, citing how that same “human approval” has famously failed to stop widespread tokushu sagi (specialized phone/wire fraud) in Japan. The article doesn’t get into this, but AI-driven vulnerability discovery is a double-edged technology usable by both attackers and defenders, so there’s also an element of a race over who masters it first. The remark about Japan having been “protected by the language barrier” also seems to reflect a recent trend: as translation AI improves in accuracy, attack techniques and scam schemes originating overseas are increasingly flowing into the Japanese-language environment too.
*This article is compiled from excerpts and a summary of the 5ch (Science News+) thread “Sam Altman: “There’s no way we can fix every single bug anymore”.”
Leave a Reply