From our other sites
The story
On October 21, OpenAI announced that its own AI model had discovered a vulnerability in the open-source language PHP, with a fix released that same day. OpenAI predicts that “similar discoveries will surge in the coming months.” On 5ch, reactions mixed welcoming AI-driven vulnerability discovery with criticism over the chaos caused when information leaked before the fix was ready. Alongside that, plenty of sarcastic calls to ditch internet dependence and go back to fax machines and hanko stamps stood out.
10/22 (Thu) 1:14 AM
ITmedia NEWS
On October 21 (local time), OpenAI announced that one of its AI models had discovered a vulnerability in the open-source programming language PHP, with a fix released the same day. The company predicts that “similar discoveries will surge in the coming months.”
Source: news.yahoo.co.jp / Original article here
What people said
You guys do the same thing — call it a 'pile-on' (enjou, Japan's term for online mob-shaming) and use whoever looks weak as a punching bag.
This mindset is exactly why Japan's security is bottom-tier.
That's probably how it's gonna be from now on.
Cue the land-fraud scammers (jimenshi, con artists who sell land they don't own) nodding: 'Tell me about it!'
Asahi Beverages actually did exactly that when they got hacked.
Kind of sad, but if the whole world went back to paper, hanko, and signatures, Japan's economic competitiveness would probably shoot right back up like it was in the early Heisei era (late 1980s–90s).
V Point's been going into maintenance every month lately.
All the engineers who couldn't spot the flaw should just quit the industry lol
It was bound to need fixing eventually — the AI just happened to find it first.
Basically, AI can define and store rules — even on the free tier.
ChatGPT's free version gets pretty sharp once you've fed it a ton of stored rules.
Gemini, not so much.
The real issue is how you write and store the rules, plus the concept of feedback.
Way too many users have no clue about any of that.
AI can fundamentally define and store rules, and even on the free tier, using it properly can significantly boost the accuracy and consistency of its answers.
But a lot of users don't understand how to create and store rules, or the feedback concept — analyzing what's wrong with an answer and feeding that back in to improve it.
Problems with the default setup:
Your personal judgment criteria and response rules aren't adequately reflected.
Past corrections and failures aren't systematically accumulated into rules that prevent recurrence.
There's no mechanism to analyze flawed answers and continuously improve them.
Leave your instructions vague, and you get answers that miss the point, unnecessary pushback, and inconsistent formatting.
With complex conditions or multiple interacting rules, since the application conditions and priority aren't sorted out, it's hard to keep answers consistent.
In short, the bigger problem isn't the AI's raw ability itself — it's the lack of accumulated rules and proper operation, meaning its true potential never gets fully drawn out.
No guarantee those investigative agencies themselves aren't doing unauthorized access too.
The fact that you can even think of that — you're an AI yourself, aren't you?
Don't worry — once we can't keep up with the AI cat-and-mouse game anymore, we'll just revert to analog.
Attacking advertises the need for defense, buyers flock in, and the defenders funnel money back to the attackers.
Apparently it's already been distributed, though.
It's become a huge problem because some idiot calling himself a security researcher went and published it.
Wild that an actual working engineer wouldn't know what happens if you publicize a hole before it's patched.
That's just the behavior of some idiot who doesn't want to let go of the spotlight. End result: the weakness just stays exposed longer. What an idiot lol
The internet is a world where you can't get anything done unless you leave the door cracked open a little. Can't be helped.
It'll go away once humans and computers both stop evolving in every sense lol
Background and Key Points of This Discussion
PHP is a language used widely across web services around the world, so any vulnerability found in it has a huge blast radius. This time, OpenAI disclosed how it found the flaw at the same moment the fix went live, but the thread also raised a separate point — that a fix was reportedly being prepared quietly, meant to be announced only after patching, only for information to leak early and cause confusion. That’s a case of information-management trouble on the developer-community side, distinct from OpenAI’s own announcement; reading the two together without separating them risks the misunderstanding that OpenAI’s announcement itself caused the chaos. The discussion split along two lines: whether to view AI-driven vulnerability discovery positively, versus jumping to claims that “human engineers are becoming unnecessary” — with the latter being something of a tangent from the main topic. The punchline about returning to fax machines and hanko stamps is best read not as an actual proposed countermeasure, but as sarcasm aimed at the risks of internet dependence.
*This article is excerpted and summarized from the 5ch (News Speed+) thread “PHP Vulnerability Discovered by OpenAI’s AI — “Similar Discoveries Will Surge in Coming Months”.”
Leave a Reply