PHP flaw: How AI just hacked an unhackable part of the internet | About That — Transcript
Full transcript
- 0:00Ah, PHP, where the P stands for PHP.
- 0:04No, really. That's actually what the
- 0:06first P stands for. But, never mind that
- 0:08because more important than this stupid
- 0:10factoid you will never now forget is
- 0:12knowing that PHP is a critical piece of
- 0:16what most of the internet is built on.
- 0:18And that a critical flaw in PHP is now
- 0:21public knowledge.
- 0:22>> There's an urgent warning tonight for
- 0:24organizations to install [music]
- 0:25critical security updates to protect
- 0:28websites around the world under threat.
- 0:30>> This is all after a major security flaw
- 0:32was found in a programming language that
- 0:34runs much of the internet.
- 0:36>> Now, I'll get into more technical detail
- 0:37in a moment, but this newly discovered
- 0:39bug in the internet found and powered by
- 0:43AI is so serious that many of the
- 0:46world's biggest companies just decided
- 0:48going offline, unplugging from the web,
- 0:51might just be the safest thing.
- 0:52Cybersecurity companies have tracked
- 0:54millions of hacking attempts now that
- 0:56this exploit is in the wild. And what we
- 0:58don't know right now is how much damage
- 1:01was done when this exploit leaked but
- 1:04before it could be fixed. In fact, there
- 1:06are almost certainly parts of the
- 1:07internet that still aren't fixed even as
- 1:10you watch this video. So, it's open
- 1:12season for hackers and a race for
- 1:15defenders in what experts are calling
- 1:17one of the most severe vulnerabilities
- 1:19of its kind, quite possibly in all of
- 1:23software history.
- 1:29PHP is just a programming language and
- 1:33organizations run it on their own web
- 1:35servers to basically fetch information
- 1:37when someone online asks for it. So, for
- 1:40example, let's say you want to know
- 1:42which is the correct orientation for a
- 1:44toilet paper roll. Paper comes out over
- 1:47or under. So, naturally, you go to
- 1:49Wikipedia to look it up. You type in
- 1:51toilet paper orientation and completely
- 1:54as expected, you get a remarkably
- 1:56detailed breakdown including a reference
- 1:59to the original toilet paper holder
- 2:00patent. PHP is how a server processes
- 2:04your request, gets you what you need,
- 2:06and then generates that webpage before
- 2:09it reaches your browser. And just to
- 2:10make my point clear here, PHP is part of
- 2:14the foundational architecture of how
- 2:17websites work at all. It's not just
- 2:20software, it's also the very language
- 2:23that such software understands. And it's
- 2:26that elemental nature of PHP, how it's
- 2:30embedded so deeply in so much, which
- 2:32begins to explain why this flaw that
- 2:34they found in PHP is estimated to affect
- 2:3870% of the internet.
- 2:40>> There are government websites, there are
- 2:42websites that are used for shopping
- 2:44enablement, WordPress uses it, [music]
- 2:46Wikipedia uses it.
- 2:47>> I struggle to think of another
- 2:49technology today that's as widespread or
- 2:51impactful and used across the planet in
- 2:55our tech stack. We all think Mac,
- 2:57Microsoft, Google, but PHP, it's the
- 3:00giant sitting below everything.
- 3:03>> It's not really an exaggeration to say
- 3:05PHP is a backbone of the web, part of
- 3:07this invisible infrastructure that makes
- 3:10it all tick. So now let me show you
- 3:12what's broken about it.
- 3:15>> [music]
- 3:19>> So the most urgent part of the story
- 3:21starts with a cybersecurity company
- 3:23called Wiz, owned by the same company
- 3:25that owns Google. They used the
- 3:27processing power of the most advanced
- 3:30large-scale artificial intelligence
- 3:32models and systems available to see if
- 3:35PHP had a weak spot. What they found,
- 3:39I'll admit, is very technical, and I can
- 3:42give you one example just to show you
- 3:44what I mean. This bit here about remote
- 3:46process memory disclosure. So it's
- 3:49talking about how PHP handles RFC 1867
- 3:54multipart requests, which are a type of
- 3:57file upload, before normal module
- 4:01initialization has completed. So,
- 4:04the server is managing the upload and
- 4:07doing a few other related things at the
- 4:08same time. Like a waiter taking orders
- 4:12before the kitchen staff are all set up.
- 4:14This is normal. And when this bug report
- 4:16talks about publishing a session ID,
- 4:19that's like the waiter writing your
- 4:21order down on a ticket.
- 4:23But, here's where it gets complicated,
- 4:25where an out-of-memory error can abort
- 4:28the request before cleanup completes.
- 4:31PHP then destroys the request heap and
- 4:35may leave the session ID pointing to the
- 4:37freed allocation, and then we have this
- 4:39bit about PHP trusts the stale pointer
- 4:43and ultimately grants unauthenticated
- 4:45disclosure about things outside the
- 4:48attacker's request.
- 4:50What the heck does that mean? Well, if
- 4:52we return to our waiter analogy with
- 4:54you, the customer, as the hacker, the
- 4:57waiter's writing down your order, but as
- 4:59the person trying to break the system,
- 5:00you just don't stop, and you somehow
- 5:04manage to bamboozle him enough with
- 5:06enough weird stuff that he's got to put
- 5:09everything down for a second, and he
- 5:10tells you, "Okay, wait. Stop. Let's
- 5:11start again. This is too much."
- 5:13But, then, in the confusion, his orders
- 5:16are getting mixed up. And as you place
- 5:18your new order, he's trying to make sure
- 5:20he gets it right, reading your new order
- 5:22back to you, except mixed in there are
- 5:25elements of your old order, and maybe
- 5:27even the orders of other customers that
- 5:30have nothing to do with you.
- 5:32His tickets,
- 5:34his very memory, is mixed up. That's the
- 5:36sting of this bug. We demonstrated
- 5:40unauthenticated disclosure of live
- 5:43worker memory containing data from
- 5:45outside the attacker's request.
- 5:48And the way experts at Wiz put it, that
- 5:50memory may contain passwords,
- 5:53authentication tokens, API keys, and
- 5:55certificates belonging to other requests
- 5:57or the server itself.
- 5:59>> So, imagine it's a a government website
- 6:02or bank website, and so now you can
- 6:04listen in and gather things like
- 6:06passwords and and usernames or even
- 6:10take control of accounts. So, it is a
- 6:13whole bunch of headaches
- 6:15all at once.
- 6:16>> In the industry, we call it god-like
- 6:18access, which is obviously the last
- 6:20thing you would ever want if you're
- 6:21running a government website or a
- 6:22financial services or a healthcare
- 6:24website or anything involving private
- 6:26data that you don't want to get out or
- 6:28systems that need to be available to
- 6:30their communities 24/7/365.
- 6:33>> And this is just one of several bugs.
- 6:35There are others, but that all fall
- 6:38under the same general weakness [music]
- 6:40of how an attacker can overload parts of
- 6:43PHP memory without needing [music] any
- 6:46kind of credential at all. And then use
- 6:48that confusion to give them access to
- 6:50things that they shouldn't be seeing.
- 6:51And on some levels, just to go back to
- 6:53the waiter analogy, these loopholes can
- 6:56even allow the attacker to throw away
- 6:59other people's orders or worse, write
- 7:02new ones for the waiter to fulfill.
- 7:04>> Imagine being [music] able to go into a
- 7:06website, going into a database, and
- 7:08putting misinformation, changing [music]
- 7:10information within a corporate corporate
- 7:13environment. That could have equally
- 7:15devastating results.
- 7:16>> In the same way we have standardized
- 7:19ways of measuring the magnitude of
- 7:21earthquakes or the power of a hurricane,
- 7:24we also have what's called the CVSS or
- 7:28the common vulnerability scoring system,
- 7:30which is basically a way of rating how
- 7:33severe a software or system
- 7:35vulnerability is. These PHP exploits
- 7:38have a severity reaching 9.2 out of 10.
- 7:42Some security firms even actually rate
- 7:43them a perfect 10 out of 10, which means
- 7:46a bad actor can crash the system or
- 7:49hijack it, wipe data, change it, steal
- 7:52[snorts] it. And exploiting this
- 7:54vulnerability was only made possible
- 7:57with Frontier AI, basically the most
- 8:00powerful AI systems available today.
- 8:02>> It was AI that allowed
- 8:06analysis of the millions of lines of
- 8:09code that make up PHP to find where that
- 8:12weakness was. If we relied on humans to
- 8:15do that kind of analysis, it would take
- 8:17hundreds of years to get to that point.
- 8:20>> Yeah, because the bug only works when
- 8:22it's chained with multiple other bugs,
- 8:25the outcomes of which can all vary
- 8:27widely depending on which PHP version
- 8:29you're running, how it's configured.
- 8:32Exploiting the bug is like solving a
- 8:35maze where every move you make changes
- 8:38where one of the walls is. Solvable, but
- 8:41maddening to do so. Just to prove that
- 8:43point, the underlying vulnerabilities
- 8:45being exploited here are so arcane and
- 8:49complex, they've apparently been there
- 8:52for decades, but nobody could figure out
- 8:54how to actually use them until
- 8:57AI.
- 8:58And now,
- 9:00there's just one more problem.
- 9:06You know, there's a way of looking at
- 9:08all of this as a good news [music]
- 9:09story, right? After all, it was the good
- 9:12guys, the defenders, who found the
- 9:14exploit and are trying to patch it up.
- 9:17It's just that they haven't quite
- 9:18finished patching it up yet. They've
- 9:21been working quietly for months, since
- 9:24the summer, to create a fix that they
- 9:25could then distribute in an orderly,
- 9:28large-scale way. Then, the plan was to
- 9:30tell the general public about it.
- 9:32Except, it all leaked before they could.
- 9:35>> So, one of the research firms that was
- 9:37doing original work on this
- 9:38vulnerability accidentally leaked a
- 9:41video that they had done of them proving
- 9:43the concept that this bug works. That
- 9:45showed people that this vulnerability
- 9:47can actually be used to get into
- 9:48systems. It's not just that it's a
- 9:50theoretical problem.
- 9:51>> Multiple stories right now about who
- 9:53leaked what. We're seeing others
- 9:55accusing some vendors of jumping the
- 9:57gun. Basically, the coordination fell
- 10:00apart on this.
- 10:01>> So, now you have a new way of
- 10:03understanding what I showed you at the
- 10:04very start. Why companies, even
- 10:07governments, have panicked pulling down
- 10:09their websites, which right now feel
- 10:11more like sitting ducks. Why hackers,
- 10:14even now with a patch available, are
- 10:16still trolling the web looking for
- 10:18targets that haven't yet gotten the
- 10:19memo. And it's why these attempts to
- 10:21break in were already happening even
- 10:24before the wider public ever caught wind
- 10:26of the story to begin with. So, it's a
- 10:29real-time race. AI being used to
- 10:32discover and close off these
- 10:34vulnerabilities, while also being used
- 10:36to hack the unhackable, to break into
- 10:39systems that no human or even team of
- 10:41humans could do before. What's unnerving
- 10:44about the whole thing though is that
- 10:46even the patch that closes this loophole
- 10:49may not be enough. It's like when
- 10:51squirrels chewed their way into my dad's
- 10:53house years ago.
- 10:54You can patch the hole, but you still
- 10:56have squirrels.
- 10:58It's impossible to know not just if
- 11:00hackers are finding their way into
- 11:02systems through this PHP exploit, but
- 11:04even what are they doing once they're
- 11:07inside? Is it a smash and grab that
- 11:09we'll know about right away? Or are they
- 11:12planting malicious code that works
- 11:14patiently? Is it possible that they're
- 11:18creating other weaknesses that we may
- 11:20only learn about much later?
- 11:25>> [music]
About this transcript
This page contains the full transcript of PHP flaw: How AI just hacked an unhackable part of the internet | About That by CBC News, generated from the public captions YouTube serves with the video. The transcript has 1,801 words across 283 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.