Video summary
Как вчера ИИ положила весь интернет
Main summary
Key takeaways
PHP Vulnerabilities and AI-Assisted Exploit Research
On October 21, 2026, patched PHP versions were released for two serious vulnerabilities. Researchers had turned the flaws into working attacks with help from neural networks, prompting urgent updates across PHP-powered services.
The incident illustrates how AI can make old software bugs more practically dangerous—not because AI independently hacked websites, but because it helped researchers analyze code, explore attack paths, and refine tests.
What the vulnerabilities involved
PHP is a server-side language whose interpreter is used by many websites, including those built with WordPress, Drupal, and Joomla. The video cites W3Techs’ estimate that PHP is used by nearly 70% of websites for which the server-side language is known. A flaw in the shared interpreter can therefore affect sites even when their developers did not write the faulty code.
The vulnerabilities involved memory handling:
- In one case, PHP could treat an incompletely initialized record as ready if processing ran out of memory at the wrong moment.
- In another, PHP freed memory but retained a reference to it. A later request could reuse that memory while the program continued to access it through the old reference.
Researchers found ways to use these conditions to execute attacker-controlled code. In some cases, they could also read server memory that might contain sensitive data. Because the attacks targeted PHP processing rather than a site’s login page, passwords and two-factor authentication would not prevent the initial compromise.
The feasibility of an attack depended on how PHP was deployed. Researchers reportedly demonstrated a route through Apache that could affect a server before the site’s code ran. With PHP-FPM, the process might instead terminate, so researchers needed a different route through an already running application.
Why the issue became urgent
The video says the underlying memory problem was noticed in 2024, but developers had not settled on a safe fix. Proposed changes risked compatibility problems, server outages, or requiring changes to extensions. Meanwhile, researchers had not yet demonstrated a reliable exploit, making it seem less urgent to delay action.
That changed in 2026. The Kali team began a new PHP review, and researchers—including the Wiz team—developed working methods for exploiting the flaws. Neural networks helped inspect the code, identify risky sequences, suggest similar locations, and analyze test results. Researchers still had to validate whether crashes could be reproduced and whether they could lead to the intended behavior.
The Wiz team reportedly demonstrated attacks involving WordPress, Drupal, and Joomla. The video emphasizes that AI can shorten the path from identifying a bug to producing a usable exploit. The difficulty of exploiting a flaw should not be assumed to remain constant.
Response and practical implications
- The vulnerabilities were rated 9.2 out of 10 in severity.
- Attack details and vulnerability identifiers reportedly leaked before the planned disclosure timeline. A demonstration video was also briefly published and copied.
- Amazon reported attempted exploitation against its honeypot servers before the patch release.
- PHP released fixed versions on October 21. Website owners and hosting administrators then had to locate affected servers, install the update, and check that their applications still worked.
- Cloudflare prepared temporary filtering protections, but the video stresses that these do not remove the underlying flaw: PHP still needs to be updated.
- Some services were temporarily taken offline as a precaution. The Netherlands reportedly shut down some government websites, including DigiD. Square Enix later said its update servers were secured and Final Fantasy XIV was available again.
The video’s main takeaway is that publishing a patch is not the same as securing the internet: hosting providers and server administrators must actually deploy it. Until then, owners may need temporary protections or to close access to affected services.
Content and sources
Guide or tutorial: This is an incident analysis, not a step-by-step guide. It explains the vulnerability mechanics and response but does not provide operational exploit instructions.
Main speaker and sources: The presenter is Roman Sakutin. The analysis refers to research from the Kali and Wiz teams, an explanation from the PHP Foundation, and reports or actions attributed to Amazon, Cloudflare, the Netherlands, and Square Enix.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.