they’re shutting it DOWN
AI-assisted researchers using OpenAI and Google models found critical remote-code-execution flaws in PHP ("PHP2Shell"), which runs an estimated 70% of websites. Governments took sites offline and attacks began. Wes Roth argues this shows AI has changed vulnerability discovery: researchers with AI exploited WordPress within two days, where a 57-day human audit in 2024 had found nothing. He says unmaintained sites and smaller organisations are most exposed, and he urges users to protect themselves.
Video1 source3 surprising claims
- Borne out by another source: 2
- Said here alone: 40
In English; the Slovenian edition is in preparation.
What it says
- Wes Roth reports that AI has uncovered critical security flaws in widely used web technology. The best known is a pair of PHP vulnerabilities that let attackers run code without logging in, for example on WordPress sites. The Dutch government and other EU governments took websites offline. An agency Roth calls the NSA advised organisations to take PHP systems offline until they are fixed. Cloudflare reports early attacks, and Roth repeats an unverified report that the details leaked early. A second problem, raised by Pash, is that the Indian certificate authority eMudhra reportedly used the number four as a private key. Roth blames government cronyism for this. He argues that smaller organisations and unmaintained websites are most at risk. Big-tech firms with access to AI security models are safer. The flaws were found by Calif under OpenAI's Patch the Planet initiative, using GPT and Gemini models and a self-improving "skill" file. They exploited WordPress within 48 hours, while a 57-day human audit in 2024 had failed. Roth recommends antivirus, browser updates, multi-factor authentication and unique passwords. He warns that more disclosures are coming.
- Wes Roth says PHP runs an estimated 70% of websites. He adds that the big PHP exploit was discovered partly thanks to OpenAI. u2 · Roth gives the 70% as an estimate.
- Wes Roth reports that hackers used Claude, Grok and Kimi to automate a cyberattack on 400 banks. u4 · The model names come from automatic captions and appear garbled.
- The Dutch government and several other EU governments shut down many of their websites as a precaution because of the critical PHP flaws. u8–12 · Roth admits he is slightly exaggerating when he says the entire Dutch government shut down.
- An advisory that Wes Roth attributes to the NSA told organisations using PHP to prepare for an update fixing two critical vulnerabilities. It recommended taking systems offline until the flaws are fully mitigated. u9–11 · Roth names the issuer as the NSA. The listed sources point to a national cybersecurity agency, so the attribution may be confused.
- The PHP vulnerabilities let an attacker, under the right conditions, execute code on WordPress sites without logging in. Roth says they often give a direct path to the server's file system or database. u9–10 ·
Overturns the expected view1 source
Also said by
- Article PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet | Wiz Blog
- According to Wes Roth, Cloudflare was already seeing early attacks exploiting the PHP flaws. u13 ·
- Wes Roth recorded his account about three hours after the PHP vulnerability news broke, so more information was expected later. u16 ·
- Cybersecurity expert Matt Johnson advises running antivirus and anti-malware on every device, recommending Bitdefender for home use. He also advises using only Safari, Firefox or DuckDuckGo, with ad blocking on and autoplay of active content off. u17–18 · Johnson has no opinion on Brave.
- Pash reports that eMudhra, an Indian certificate authority trusted by every browser, was supposed to use 116-digit keys but used the number four. Pash says its certificates were probably compromised for about 18 months. u24–26 · Roth says he cannot judge this himself and passes on Pash's account.
Goes against what one would expect
- The security company Calif, which found the PHP flaws, joined OpenAI's Patch the Planet initiative on July 30, 2026. The initiative aims to patch critical software before attackers can exploit it. u32–33 · Captions garble the company's name; the listed source is calif.io.
- Calif's researchers used OpenAI and Gemini models, which Roth names as GPT 5.6 and Gemini 4. The researchers targeted interrupted work during PHP's cleanup after out-of-memory errors, which is where the vulnerability lies. u34–35 · The model names are garbled in the captions ('Gemini 4 argan', 'GPT 5.6 six soul').
- A 2024 audit of the PHP code, which Roth believes humans did largely by hand, lasted 57 days and did not find the flaw. u44–46 · Roth only infers that the 2024 audit was done by hand ('likely').
- Wes Roth gives the PHP2Shell timeline: researchers found the first PHP vulnerability on August 19. They achieved remote code execution against WordPress within 48 hours, reported the flaws to the PHP security team, and disclosed them publicly on October 21. u47 ·
1 source
Also said by
- Article PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet | Wiz Blog
- Wes Roth recommends multi-factor authentication instead of passwords alone, and says hardware authentication methods are probably better than text messages. u19 ·
- Wes Roth suggests updating passwords now and using a unique password for every account. That way one leaked password does not open all of a person's other accounts. u21–22 · For ordinary users whose data may already be on unmaintained sites.
- Wes Roth advises anyone who runs a website to make sure it is updated and to get confirmation that the fix is actually running. u22 · For website owners.
- Citing rumours from Anthropic, Wes Roth predicts more vulnerability disclosures soon. He compares the moment to February 2020, just before the COVID lockdowns. u49 ·
- Wes Roth explains that a compromised but legitimate-looking site, even one showing the security icon, can harvest personal data and then forward the visitor to the real site. The victim may never know. u7–8 · A trusted domain name and a padlock are no guarantee of safety.
- Wes Roth argues that the greatest danger lies in websites that hold private data but are no longer maintained. Nobody will apply the patch, so a large part of the web stays exploitable after the fix ships. u12–13 · Because of this, the risk lasts long after the patch is released.
- Pash and Wes Roth attribute eMudhra's weak keys to incompetence protected by political connections. A firm that secures government contracts through relationships, or possibly bribery, need not be good at its work. u26–28 · Pash concludes that anything not run by big tech in such countries is suspect.
- Wes Roth compares the PHP flaw to context compaction in chatbots like ChatGPT. Something can go wrong at the point where a system is interrupted to clean up, just as models can lose key instructions when they summarise their context. u36–37 · The analogy makes a technical bug class easier for non-experts to grasp.
- Researcher Austin had ChatGPT turn the first exploit into a reusable skill.md file of instructions for finding similar issues. Each new discovery improved the skill, much like Andrej Karpathy's autoresearch loop. u38–41 · It shows how AI research compounds: skills improve themselves with each finding.
- Wes Roth explains that AI multiplies the paths a researcher can explore. Where a human tests ideas one by one, a language model can take the researcher's accumulated knowledge and test many paths at once. u41–42 · Roth uses this to explain why AI collapsed discovery time from months to days.
- Wes Roth says AI has uncovered critical flaws in PHP and other web technology that put much of the internet at risk. He says people should be extremely careful, or stay off the web, for the next few days. u1–2 ·
Goes against what one would expect
- Wes Roth says his earlier warnings to secure accounts, given when Anthropic's Mythos model went to trusted partners, have been borne out. He rejects critics who called those warnings a marketing stunt. u5–6 ·
- Wes Roth urges ordinary users to protect themselves now. He recommends extra caution with familiar sites, updated software, antivirus, multi-factor authentication and new, unique passwords. u16–22 ·
- Wes Roth argues that unmaintained websites and smaller organisations are most at risk from the PHP flaws. Big-tech firms and members of AI security programmes such as Project Glasswing are much safer. u29–31 ·
- Wes Roth argues that AI-assisted research has changed vulnerability discovery. People who call it nothing new are missing the point, because researchers with AI did in two days what a 57-day human audit failed to do. u44–48 ·
- Is AI-assisted vulnerability research really nothing new, as critics say, since humans could do it too? u43–48 · Roth answers no: humans had 57 days and did not find the flaw, while researchers with AI did it in two days.
- Will the internet have to be locked down, reversing 2020, when people were locked at home and went online? u50–51 ·
- Wes Roth addresses viewers who are surprised by the AI-driven security crisis. u6 · It shows Roth's view that the crisis was foreseeable.
- Wes Roth sums up how fragile modern digital architecture is, given the private key that was the number four. u11 · It captures the absurd weakness behind web trust in one line.
- Wes Roth quotes a TL;DR that he says captures the whole situation. u15 · Roth endorses it as a succinct and accurate summary of the risk.
- Wes Roth answers the claim that humans could have found the PHP flaws without AI. u46 · It states the core of Roth's argument that AI changes the game in practice.
- Wes Roth says security researchers recommend that people stay off the web for a couple of days, or at least strictly limit what they do online. u2 · Normally the advice would be to patch and carry on, not to avoid the web altogether.
- Some security software used the number four as a private key, where a secure key would be a random number about 115 to 116 decimal digits long. u3–4 · One would expect a trusted security provider to use long random keys, not a single digit that a naive count reaches in four guesses.
- Wes Roth says he is not a security expert or researcher, so his safety tips are offered only to the best of his ability. u7 · This limits how far his advice can be relied on; he partly relies on expert Matt Johnson.
- Wes Roth could not verify the report that a security researcher leaked the PHP vulnerability details before the official release. He calls the situation live and uncertain. u13–14 · An early leak would explain why attacks began before patches could be applied.
- Wes Roth notes the irony of telling viewers not to click links while offering links to sources about the hack. u22–23 · Roth leaves it to viewers to decide whether to trust the links.
- Wes Roth is uneasy that safety from AI-era attacks is concentrated in big tech and AI insiders. He says this favours the powerful over the underdogs and lets power and money flow upward. u31–32 · At stake is whether AI security tools widen the gap between large and small organisations; Roth leaves it unresolved.
- AI-assisted vulnerability research has changed cybersecurity. The PHP2Shell flaws found within days by researchers using GPT and Gemini put the unmaintained web at serious risk, so users and website owners need to act now. What happened with the AI-discovered PHP2Shell vulnerabilities, why does AI-driven discovery matter, and how can people protect themselves?
How it argues
- thesis AI-found PHP flaws endanger the unmaintained web What happened with the AI-discovered PHP2Shell vulnerabilities, why does AI-driven discovery matter, and how can people protect themselves?
- theme PHP2Shell threatens much of the web Critical PHP flaws allow code execution on WordPress sites without logging in, and governments, agencies and Cloudflare are responding to an urgent, unfolding threat.
- idea AI-found PHP flaws put the internet at risk
- evidence PHP runs about 70% of websites
- mechanism Code execution on WordPress without logging in
- evidence Dutch and EU governments shut down websites
- evidence NSA-attributed advisory recommends taking systems offline
- evidence Cloudflare already sees early exploit attempts
- surprise Researchers advise staying off the web
- quote A TL;DR that captures the situation
- idea PHP2Shell timeline from discovery to disclosure
- idea AI-found PHP flaws put the internet at risk
- theme AI has changed vulnerability discovery Researchers using GPT and Gemini found in days what a 57-day human audit missed, because AI explores many paths at once and builds reusable skills.
- theme The unmaintained web bears the risk Unpatched sites and smaller organisations stay exposed while big tech and AI insiders are protected, which concentrates safety among the powerful.
- theme Security foundations are shockingly fragile A trusted certificate authority used the number four as a private key, which shows how brittle digital infrastructure is.
- theme What users and site owners should do Ordinary users should take concrete protective steps now, and site owners should confirm their patches are running, though Roth's advice comes from a non-expert.
- theme More AI-driven disclosures are coming Roth's earlier warnings have proved right, and he expects more disclosures, comparing the moment to the eve of the 2020 lockdowns.
- theme PHP2Shell threatens much of the web Critical PHP flaws allow code execution on WordPress sites without logging in, and governments, agencies and Cloudflare are responding to an urgent, unfolding threat.
The source, passage by passage
- u1 Well, the entire internet is on fire. I mean, not literally, but you should be extremely careful if you're browsing the web, especially for the next couple of days. AI has discovered massive security flaws across some of the biggest programming languages that are used to run the web.
- u2 For example, one of them is PHP runs an estimated 70% of websites. Security researchers are recommending that people stay off the web for a couple days so they don't get spawned or at least really limit their activities. By the way, the big PHP exploit was discovered in part thanks to OpenAI.
- u3 That's the one that's getting a lot of attention, but it's not the only one. Just to give you a sense of how crazy some of these things are, a secure private key might be a randomly generated number that's 115 to 116 decimal digits long. on screen I have an example in hexadimal so that's one number split across three lines right so it's it's a very big complicated number that would be very very hard to guess some security software and coding languages they used four and I don't mean like four digits long I mean the number four I'm not kidding this is not a joke one of the private keys was literally the number four so if you had something that was just randomly going through all the numbers
- u4 starting at one. It would go 1 2 3 4 and that's it. There you are. You guessed it. Have all of my digital stuff. Meanwhile, hackers used Claude Grock and Kimmy to automate a cyber attack on 400 banks. Automated cyber attacks are becoming a norm.
- u5 Now, if you've been following this channel, we've been talking about this and expecting this months ago when Mythos it didn't come out, but it got released to some trusted partners. There were people like myself telling you to secure your stuff or you're going to have a bad time.
- u6 And there was also some other people that were saying, "Nah, that's just a marketing stunt. Nothing is changing. It's all just fugazi fugazi. It's nonsense." Well, here we are. I really hope you're not surprised by this because if you're surprised, you're not paying attention. So, in this video, let's briefly go over what exactly happened.
- u7 I'll give you some tips to stay safe to the best of my ability. I'm not a security expert, security researcher, but like many of you, hopefully you're doing this as well. I'm trying to learn more about this stuff because you could be on what you think is IRS.gov and it's got the little security icon next to it, but it's some bad guy, bad person harvesting your information and basically covering up their tracks and be forwarding you to the IRS.gov, the actual website afterwards.
- u8 But meanwhile, they have your social, they have your bank details. You may never know. The entire Dutch government has shut down. I'm just slightly exaggerating, but they've shut down a lot of their services. And that's not an overreaction. Here's the NSA, like the people that watch you when you sleep, that the NSA, right?
- u9 They're saying effective immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide direct path to the service file system or database. So basically without logging in you can run a lot of the commands.
- u10 So like if you have WordPress WordPress is is massive runs a massive amount of it runs a massive amounts of content websites online that uses PHP. So normally you log in you can do stuff with these vulnerabilities under the right conditions the hacker the attacker can just execute code without logging in.
- u11 And the NSA is saying it's recommended that organizations take the system offline until it can be fully mitigated. This this is excellent. So this is all modern digital architecture. The entirety of the internet everything everything it all relies on no one guessing the number four levels IO's disconnecting all the sites VPS servers from the web.
- u12 and a number of EU governments are shutting down their websites because of these critical flaws, specifically in PHP. The problem is it's very likely that there's a lot of websites out there that are quote unquote secure, so they have access to some private important information, but they may not be maintained.
- u13 There might not be somebody awake at the wheel sort of ready to hit the update button when it goes live. Cloudflare is already seeing early attacks. It seemed like one of the cyber security researchers did just post this information online, like it was leaked before the fact.
- u14 Now, I couldn't verify this again. This is all kind of happening live. So, take that with a grain of salt, but it does seem that this was leaked ahead of time. And here's the TLDDR that I think really sort of encompasses the entirety of what's happening in a very succinct and accurate way.
- u15 a large portion of the unmaintained internet is about to be pawned and god help us all or something like that. All right, with all that said, let's break down what happened and how to make sure that you're safe. So, really fast, let's talk about how you can protect yourself, what steps you should be taking right now.
- u16 Keep in mind this is 3 hours after this whole thing broke. So, we'll probably have a lot more insights as time goes along. But for the time being, understand that familiar websites could be taken over and used against you, against the visitor. So, just because you're visiting a familiar name, that does not mean that it's clean.
- u17 Right now, there's this massive scramble to patch everything up to just make sure that everything's good. So, especially for the next few days, be extra careful. Maybe don't go surfing the web unless you have to. Here's Matt Johnson. So he's kind of giving some of his own advice. He's saying make sure you're running full service anti virus anti-malware on every device.
- u18 He's a cyber security expert by the way. So all that is including phones. He's recommending Bit Defender for home/personal use. Update your browser apps and use only Safari, Firefox or DuckDuck Go. He has no opinion about Brave. Make sure ad blocking is turned on and autoplay of active content is turned off.
- u19 Be careful of any links you receive in text messages, email, etc. Keep your operating system and applications up to date and keep watching for new updates. Have two-factor authentication or multifactor authentication. So, you know, don't just log in with a password. Make sure you're getting a text message or if you're using those hardware authentication methods, that's probably even better.
- u20 But I think the real thing to understand is that there's a lot of websites out there where you might have logged in, use your password, and a lot of them are not being updated. Nobody's running them. Point being is like not everyone will be on top of these news and taking steps to make sure that everything's safe.
- u21 So, it does seem like a lot of your information out there could have been kind of retroactively like if you've already put it out there, it might already be lost. So, this might be a great time to think about maybe uh updating your passwords. Ideally, we'd all be having a unique password for every different application we're using.
- u22 So, every account has its very own password. So, even if one password gets lost, that doesn't mean that now the hackers are able to get into all of your other stuff. If you have a website, make sure that it gets updated and then get a confirmation that that fix is running. I'll leave some links down below so you can verify for yourself.
- u23 Although, this is kind of a weird time to be saying that. It's like don't click any links, but also here's some links to click. And here's a post by Pash. So, he's a computer scientist, founder. He's the guy on the Cognitive Revolution podcast. So he's posting some stuff about the other big kind of hack/vulnerability that we're talking about which is the e mudra.
- u24 So he's saying for those of you that are not aware of this it's an Indian digital search provider and aert a certification basically lets you know that the website that you're visiting it's real it belongs to the person that it's supposed to. So they use their relationships maybe bribery to become the provider to Indian government bodies.
- u25 Same thing for Indian state-owned banks and companies as well as some South Asian and Middle Eastern banks and institutions and I'm not social proof to become one of the most trusted authority to the global system built into every browser. So they were supposed to have password that are 116 digits long.
- u26 So very difficult to guess. They instead used passwords such as four the number four. Those certifications have been compromised probably for around 18 months. Now why is that happening? Well incompetence. These people are I mean this is Pash that's talking about this so I don't know too much about this but certainly kind of like this makes sense.
- u27 This happens in in certain parts of the world. You know after the collapse of the Soviet Union this was very common place in Russia. Basically if your business has the right connections to the government you don't need to be that good at what you do. Your contracts are guaranteed. Your competition is outlawed.
- u28 Basically it's a legalized shenanigans. And as Pash is saying here, anything not run by big tech in those countries is suspect. And this is an important point to understand. I've been talking about something very similar here. And I hate the fact that this is happening.
- u29 But right now, the places that going to be the most vulnerable will be smaller businesses. Notice currently a lot of the hacks that are happening, they're happening to certain Korean banks, certain South American banks. Chances are the large American banks for example are a lot more safer than some smaller European bank.
- u30 Companies that are in big tech like Google for example is probably a lot safer than something that's smaller or medium-sized. We're not talking about small businesses here. We're talking about like not gargantuan businesses. The companies and governments that are in Project Glass Wing, right? the ones that have access to Mythos and the analogous project by OpenAI, they're much more safer than the companies that are outside of it.
- u31 So, unfortunately, you're probably a lot more safer with a big established company, part of the big tech, part of that sort of in crowd when it comes to AI cyber security models, then you are working with somebody who's not in that inner circle. I hate saying that because obviously we want kind of like everyone else to have a chance, the underdogs to succeed.
- u32 We don't want the power and money and everything just flowing upwards. Now the people that discovered this they khif kh kh kh kh kh kh kh kh kh kh khif.io IO it's a cyber security company and they have this kind of epic looking write up of everything that happened that they made a comic book out of it. The efforts were part of patch the planet this initiative by openai after all of us kind of woke up to the fact that these AI models can find vulnerabilities and that there's tons and tons of vulnerabilities everywhere.
- u33 Vulnerabilities are the land as they used to say. the patch the planet initiative was there to basically try to patch as much of the critical stuff as possible before the bad guys had a chance to you know start hacking stuff. So they joined the initiative July 30th 2026 and then they got to working hacking PHP.
- u34 By the way they worked on this before this sort of writeup starts in 2004. this kind of insanely well done. I got to say they used OpenAI model and some of the Gemini models for this and specifically they targeted interrupted work. So a lot of this has to do with kind of out of memory corruption.
- u35 So when PHP is doing something and it runs out of memory, it has to do a cleanup process. That's kind of where this vulnerability exists. So, as we say here, we managed to surface the first out of memory fatal and the first of the three interruption points that allowed us to piece together a remote code execution against WordPress.
- u36 So, like if you're working with Chad PT, you know how if you're running it too long, it goes up. We got to like compact the context a few minutes, right? So, it's like we've reached the end of the context window. So, it takes a few minutes to like summarize everything. And often when you hear of a Chad GPT or some other model doing something horrible, it's usually because of that compaction.
- u37 it does if it forgets some key detail or some key instruction you gave it and that's when things go wrong. Well, that's a little bit kind of analogous to what happens with this PHP. So, they found this interruption point at which they able to do this rce against WordPress. So, they're able to execute some bad code that you're not supposed to.
- u38 They're supposed to remotely execute it on that machine. So, basically to hack it. So after they figured out the the first kind of example of this this researcher Austin he kind of used that idea to tell CHBT like okay so kind of generalize what we did here into a skill.mmd.
- u39 So create a skill for yourself about kind of like think about how we did this in general and kind of make a skill out of that. So they were basically reusable instructions for searching for similar types of issues. So we've talked about kind of these self-improving skills before. Andre Karpathy was the first person to kind of like really put it out there on a massive scale with his auto research project I believe it was called.
- u40 So basically having the AI run in a loop like it tests something if it works better that's the new skill. If it doesn't then it just like ignores what it tried but it keeps testing these hypotheses and every time it's able to improve its abilities it then uses that new sort of skill set or new approach. So kind of it seems like a similar idea was done here.
- u41 Each new discovery about these vulnerabilities improved that skill that the model could then use to find more vulnerabilities. And of course, AI can amplify how many different paths that the researcher can take. If it's just a human being going through these one by one, it takes time a large language model can take.
- u42 That's accumulated knowledge of the researcher and just go off on a lot of different paths and test them all. Now, I haven't read this whole thing, so I'm kind of skipping a little bit over it. Kind of glossing over the details. Again, I'll leave the link if you trust links nowadays. I'll leave a link down below.
- u43 It would be ironic to get hacked by following a link that tells you about how the hack was discovered so you could learn more about it. If that was the link that got you hacked, that would be ironic, right? Is is that the right meaning of the word? Now, here's kind of another very important point because when these AI models were just coming around getting good at doing this sort of research, people dismissed and said, "Oh, humans have been doing that.
- u44 Humans can do it. This is nothing new. Not a big [clears throat] deal. Don't worry about it. Here's the thing. We had some ideas that there might be a vulnerability here. This 2024 audit, it lasted 57 days. So, keep in mind this was 2024. We had large language models, but they weren't kind of at this level of sophistication.
- u45 So, this was likely done kind of old school ways by hand, by humans, if you will. So, it's not like we weren't aware that there was an issue. It's not like we didn't try very carefully for almost two months to look into it and find that issue. So this idea that these AI models that it's nothing new, nothing to worry about it.
- u46 The game hasn't changed at all. Humans can do all of it, blah blah blah, that's nonsense. Sure, maybe technically humans could have done it, but here's the thing. They they didn't they didn't even though they had time. They had, you know, 57 days. Now with these AI models that are assisting researchers, right?
- u47 So Gemini 4 argan and GPT 5.6 six soul. Those were the two big models that we all know that were used here, right? So August 19th, they discovered the first PHP vulnerability. Then they do RCE, remote code execution. They hack WordPress within 48 hours. Then they report these vulnerabilities to the PHP security team and then public disclosure October 21st.
- u48 So again, humans took 57 days, couldn't figure it out. Humans with these large language models 2 days and they hacked everything. So again, the people that are saying that this is just a normal technology and humans are still far far superior and this is nothing new that they're missing the plot.
- u49 And by the way, this isn't over yet because by some of the rumors from Anthropic, there's more coming soon. So don't get too relaxed. This does feel like maybe, you know, kind of early 2020, let's say February 2020, when things were kind of beginning to pick up steam kind of before the lockdowns.
- u50 Are we going to have to lock down the internet? Is this going to be the exact reverse of, you know, what happened in 2020? Back then, everybody got locked inside their homes and everybody went on the internet. What if now the internet's going to get locked down and we're all going to have to go outside again? Maybe it won't be that bad.
- u51 In fact, it might be kind of nice. But whatever the case is, be ready. If you haven't taken your cyber security very seriously and maybe you have some pictures somewhere that uh you don't want out there for the entire world to see, I don't know, maybe this would be a good time to, you know, think about patching things up a little bit.
- u52 If you want a great tip to create really strong good passwords, here's the approach that I would use. Don't just put in the number four. So using the number four and nothing else as your password is bad.
- u53 That concludes this public service announcement. Thank you for watching. My name is Wes Roth. I'll see you in the next
Around it
Odkrivanje ranljivosti s pomočjo umetne inteligence
Uporaba modelov umetne inteligence za iskanje in izkoriščanje varnostnih ranljivosti v programski opremi.
Topic2 sourcesNo primary research yet
PHP2Shell
Par kritičnih ranljivosti pomnilniške varnosti v pogonu PHP, ki omogočata oddaljeno izvajanje kode brez avtentikacije in sta bili razkriti ter odpravljeni oktobra 2026.
Topic2 sources2 surprising claimsNo primary research yet
PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet | Wiz Blog
Wiz Research reports PHP2Shell, a set of memory-safety vulnerabilities in the core PHP engine that can let unauthenticated attackers execute code on servers or read sensitive process memory. Nearly every PHP version back to 5.4 is affected, and PHP released a patch on October 21, 2026. The authors argue that frontier AI models made these longstanding flaws practical to exploit, which they present as an inflection point for the AI era.
Article1 source
Patch the Planet: global response coordination as agents uncover critical vulnerabilities
OpenAI describes its coalition response to critical vulnerabilities in PHP, which frontier AI found before attackers did, and the framework it proposes for coordinating responses to such flaws in widely used software. The page sets out a five-stage response, the support it offers maintainers and operators, and the open questions it wants the security community to help answer.
Article
Security advisory: Critical remote code execution vulnerabilities, update now
PHP's security advisory announces fixed releases 8.5.12, 8.4.27, 8.3.36 and 8.2.35 for two critical vulnerabilities that can allow unauthenticated remote code execution. It urges all users to update immediately, describes both flaws, and lists interim configuration measures for those who cannot update yet, stressing that only an update fully fixes the problem.
Article1 source2 independent witnesses
PHP
Odprtokodni skriptni jezik na strani strežnika, ki se široko uporablja za izdelavo spletišč in sistemov za upravljanje vsebin.
Topic5 sources2 independent witnesses1 surprising claimNo primary research yet
WordPress.org
Article
web-dedbc374478f
Article
Calif
A security company that discovers software vulnerabilities, including the PHP flaws it found using AI models.
Topic2 sourcesNo primary research yet
Patch the Planet
Pobuda organizacije OpenAI za iskanje in odpravljanje kritičnih ranljivosti v programski opremi, preden jih lahko napadalci izkoristijo.
Topic2 sourcesNo primary research yet
- GPT
- Gemini
- Government Office for Information Security
- Oddaljeno izvajanje kode
- OpenAI
- Po odkritih ranljivostih v PHP-ju visoka ogroženost kibernetske varnosti v Sloveniji
This text is written with AI.
Placed in: danes