TL;DR
Get privacy and security gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A report by Drew Silcock describes two critical PHP vulnerabilities, a disrupted disclosure timeline and exploitation before patches were available. The account raises concerns that AI-assisted vulnerability research may shorten the time defenders have to respond, but the scale of exposure and the role of AI in attacks remain uncertain.
A report by Drew Silcock describes two severe PHP vulnerabilities, a disclosure process disrupted by apparent leaks and exploitation before patches were available. The episode raises a pressing question for organizations running PHP: whether defenders can still patch in time when vulnerability details and working exploits may spread within hours.
The report identifies the flaws as CVE-2026-109618 and CVE-2026-109619, which it says were remote-code-execution vulnerabilities. Silcock writes that researchers at Calif and Wiz found the issues using frontier AI models. The report names the models as GPT-5.6 Sol and Gemini 4 Argon; those product and model details are not independently established by the material provided here.
According to Silcock, one issue had been identified during a 2024 PHP Foundation security audit, but the report says a fix was not agreed on because researchers could not demonstrate a practical exploit and other concerns took priority. Silcock says the PHP Group moved the planned patch date forward after details appeared on Hacker News and other information was apparently disclosed. The account describes several events unfolding over roughly 17 hours, though it does not provide a fully reconciled timeline.
The report says fixes were published before the planned end of the disclosure embargo and that the vulnerabilities were being exploited by the time patches became available. It attributes criticism of the disclosure process to the PHP Foundation, which reportedly cited a Calif exploit video and fixes published by Chainguard before the embargo expired. The source does not provide incident data or independently verified details about the attacks.
The shrinking window to patch
The immediate concern is the time available to defend exposed systems. If attackers can turn vulnerability descriptions or newly released patches into working exploits quickly, organizations may have less time to identify affected services, test fixes and deploy them. Silcock argues that AI tools could speed parts of that process, though the report does not establish that AI alone caused the attacks it describes.
That timing matters beyond PHP. Software maintainers, security researchers and organizations all depend on coordinated disclosure: researchers share enough information to prepare a fix, while limiting what attackers can use before users can update. An early leak or premature publication of a fix can disrupt that balance. For readers, the practical stakes include the security of websites and online services that rely on vulnerable software, not just the technology debate over AI.
The report also cautions against treating every online exploit demonstration as proof. Silcock cites security commentator LiveOverflow’s observation that AI can produce non-working exploit attempts, or “slop.” That makes expert validation more important: a flood of plausible but faulty claims can distract responders, while one effective exploit may still pose a real risk.
As an affiliate, we earn on qualifying purchases.
How the PHP disclosure unfolded
Silcock’s account links the incident to a coordinated disclosure process that appears to have gone off schedule. The PHP Group had planned to release fixes on a Thursday, according to the report, but advanced the announcement to Wednesday after vulnerability information surfaced. The patch itself was reportedly not available until the following day, leaving a period in which administrators knew of the risk but could not yet install the official fix.
The report says early public discussion included a maximum-severity rating, while the vulnerabilities’ actual scores were lower. It also says government advisories in the United Kingdom, European Union, Japan and United States contributed to alarm. Those advisories and their precise wording are not reproduced in the supplied source, so the characterization of them as overreactions is Silcock’s assessment rather than an independently verified finding.
The author estimates that the vulnerabilities could affect a large share of the internet and names services and institutions said to be exposed. That estimate is not supported with a measurement method in the supplied material. It should not be read as confirmation that those organizations were compromised or that 70% of websites were vulnerable.
“The time from “hey I heard there’s a bad RCE in this project” to “there are sophisticated, automated, and concerted attacks” has gone from weeks/months to hours/minutes.”
— Drew Silcock, author of the report
small business firewall appliance intrusion prevention
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Exposure and attack evidence
Several key points remain unverified in the supplied material: the number of affected systems, which named services were vulnerable or compromised, and how many attacks used these specific flaws. The report gives a broad estimate of potential exposure but does not explain its basis. It also does not provide technical advisories, patch links or incident-response data that would allow those claims to be checked here.
The precise disclosure sequence is also unsettled. Silcock describes apparent leaks, an accelerated announcement, early publication of fixes and exploitation before patches were available, but the account does not establish exactly when each event occurred or who first made exploit details public. Nor does it prove that AI-generated exploits were used in the observed attacks. The report supports a concern about compressed response times, not a definitive conclusion that AI has changed cybersecurity everywhere.
encrypted network attached storage for business backups
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Patch status and timeline checks
The next practical step for organizations is to consult official PHP security advisories and their software vendors for affected versions and available fixes. Administrators should establish whether their systems run the relevant PHP components, apply verified updates where applicable and review reliable security guidance. The source material does not identify specific versions or provide a patch procedure, so readers should rely on current vendor documentation rather than assumptions drawn from the report.
Further reporting should clarify the disclosure timeline, the technical scope of both vulnerabilities and whether attacks were confirmed against particular organizations. Statements from the PHP Group, the researchers and affected service operators could also establish how the early disclosures happened and whether the reported incidents involved AI-assisted exploitation. Until those details are available, the episode is best understood as a warning about patch coordination and response speed, not proof of a single new cyber norm.
cybersecurity incident response tabletop exercise kit
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened in the PHP incident?
Silcock reports that two PHP remote-code-execution vulnerabilities were disclosed amid apparent leaks and a changed patch schedule. The report says exploitation was underway before fixes became available, but the supplied material does not independently verify attack details.
Did AI discover or exploit the vulnerabilities?
The report says teams at Calif and Wiz used frontier AI models while researching the flaws and building exploits. It does not establish that AI was used in attacks against affected systems or that AI was solely responsible for finding either vulnerability.
Were most websites affected?
The report estimates that the vulnerabilities could affect more than 70% of the internet, but it provides no method for that estimate. That figure does not confirm that 70% of websites were vulnerable or compromised.
What should organizations do now?
Organizations should check official PHP and vendor advisories for affected versions and confirmed fixes, then follow their established patch and incident-response procedures. The source does not specify affected versions or provide verified remediation instructions.
Does this prove AI has changed cybersecurity?
No. The report describes a case that may illustrate faster vulnerability research and a compressed patching window, but it does not establish a broad, measured change across the cybersecurity industry.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
