Skip to content
AI ConnectPowered by VELENTIS
AI-generated2 min

Critical PHP Flaws: Reasoning Model GPT-5.6 Sol Uncovers 14-Year-Old Memory Bugs

Two critical PHP vulnerabilities uncovered with the help of the GPT-5.6 Sol reasoning model threaten 70 percent of web servers, triggering millions of automated attacks.

This article was AI-generated and published automatically. Context, labelling and all sources at the end of the article.

(KI-generiertes Symbolbild: Gemini / AI Connect)

On October 21, 2026, cybersecurity researchers publicly disclosed two critical vulnerabilities residing within the C core of the PHP scripting language. Cataloged as CVE-2026-109618 and CVE-2026-109619 with a CVSS severity rating of 9.2, the flaws grant remote attackers the ability to execute arbitrary code. The vulnerabilities span all PHP versions from 5.4 through 8.5, placing an estimated 70 percent of all web servers across the globe at immediate risk. The disclosure prompted urgent worldwide security notices and coordinated incident response measures by hosting providers and government bodies.

The underlying memory corruption bugs, specifically out-of-memory and use-after-free conditions, had remained hidden inside the codebase for 14 years. Their discovery was achieved not by conventional static analysis alone, but by security researchers from Calif and Wiz utilizing the reasoning model GPT-5.6 Sol. The research teams deployed the artificial intelligence system to systematically audit complex execution paths and legacy memory handling logic across the C implementation. The breakthrough highlights the capacity of advanced reasoning models to identify severe flaws that escaped human scrutiny for well over a decade.

The public release of vulnerability details immediately triggered an aggressive wave of automated exploitation routines across global internet traffic. Matthew Prince, the chief executive officer of Cloudflare, shared telemetry demonstrating an unprecedented escalation of hostile traffic. Cloudflare detected thousands of scanning attempts prior to the formal disclosure, which escalated to tens of thousands within 45 minutes of the announcement. By October 22, 2026, the volume had expanded into millions of automated attack routines attempting to locate vulnerable web deployments.

In response to the imminent threat of host compromise, authorities in multiple nations proactively took critical online government portals offline. Simultaneously, the PHP core development community issued emergency point releases to protect all actively maintained software branches. Security patches were rolled out for PHP 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to neutralize the underlying memory flaws. System administrators worldwide were urged to deploy these updates immediately or isolate legacy installations behind stringent perimeter defenses.

The incident represents a watershed moment for internet infrastructure security, a point echoed in observations by software engineer Armin Ronacher. Autonomous AI agents and reasoning-capable models are now systematically analyzing open-source foundations, exposing legacy C flaws that lingered unnoticed across generations of software releases. For network defenders, this development shifts the cybersecurity landscape into an asymmetrical race against machine-speed exploitation. The window between an automated zero-day discovery and global internet-wide scans has effectively compressed into minutes.

What this means for you

For systems engineers and technical leadership, this incident demonstrates that relying on the perceived stability of mature open-source components is no longer viable. As reasoning models make deep vulnerability research cheap and scalable, patch deployment cycles must be compressed from weeks to hours. Organizations must integrate AI-assisted static and dynamic auditing directly into their code validation pipelines before external actors find remaining legacy flaws.

Perspectives

Coverage: 3× Other

One story, several angles: how each source frames the topic, each with a verbatim quote.

  • sumsub.comOther

    Sumsub focuses on the operational fallout of the PHP2Shell vulnerabilities, highlighting widespread website shutdowns, millions of exploitation attempts, and the risks of faster AI-assisted vulnerability discovery.

    Original quote

    „The incident raises much broader concerns about the growing speed of AI-assisted vulnerability discovery and exploitation.“

    sumsub.com
  • en.wikipedia.orgOther

    Wikipedia provides an encyclopedic and factual account, focusing on the technical identifiers of the vulnerabilities, the affected PHP versions dating back to 2012, and the emergency patch timeline.

    Original quote

    „PHP2Shell is a pair of high-impact vulnerabilities“

    en.wikipedia.org
  • clarin.comOther

    Clarín frames the discovery as a historic milestone, emphasizing the race of autonomous AI models like GPT-5.6 Sol that uncovered a 14-year-old flaw affecting the majority of the internet.

    Original quote

    „What set this discovery apart was speed and analytical depth.“

    clarin.com

Source classification is maintained editorially (political spectrum only where consensus is broad; vendor communication is PR, not journalism). Unlabelled sources are unclassified: we do not guess.

Evidence

Well sourced
78/100
  • Vulnerabilities CVE-2026-109618 and CVE-2026-109619 hold a CVSS score of 9.2, affect PHP 5.4 to 8.5, and threaten approximately 70 percent of web servers globally.

    single source
  • Researchers from Calif and Wiz uncovered the 14-year-old memory management bugs using the reasoning model GPT-5.6 Sol.

    verified
  • Cloudflare CEO Matthew Prince recorded thousands of attack attempts prior to disclosure, tens of thousands within 45 minutes, and millions by October 22, 2026.

    verified
  • Emergency patches resolving the vulnerabilities were issued for PHP versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12.

    verified

The evidence score is computed, not hand-set: from confidence, the number of sources and the share of verified statements.

Source & transparency

As of: October 22, 2026

AI-generatedAI-generated: produced automatically from vetted sources with technical quality checks (source, quote and figure verification); no human sign-off of each item before publication

Sources
3
Verified statements
3 / 4
Evidence score
78Well sourced

Want to put this into practice?

We connect you with suitable AI providers from the DACH region, free of charge and without obligation.

What's next?