Chainguard co-founder and CTO Matt Moore took the stage at KubeCon + CloudNativeCon North America 2026 to justify the company's decision to ship patched PHP images ahead of an agreed embargo window. On October 21, the company released its fixes more than six hours before the coordinated security disclosure was scheduled, prompting PHP maintainers to accuse Chainguard of handing attackers an advantage. Moore's response: the advantage had already shifted to the adversaries.

In remarks to The New Stack, Moore stated, "We own the decision that we chose to do it. Nobody could force us to do it — but we did not do it in a vacuum." He pointed to discussions with other embargo participants and reports from security vendors documenting active exploitation attempts against the vulnerabilities before Chainguard's release.
Chainguard's position was not that it patched faster and therefore shipped first. Rather, according to Moore, the company assessed that attackers could already exploit the bugs or were close to doing so. The two flaws—CVE-2026-109618 (out-of-memory corruption) and CVE-2026-109619 (stale global state), collectively dubbed PHP2Shell—represented serious memory-management and error-handling defects that could allow unauthenticated attackers to execute code on vulnerable PHP servers.
The nature of the vulnerabilities
The two bugs stemmed from PHP's handling of memory exhaustion and its failure to clear references to freed memory between requests. When memory limits were reached at the wrong moment, PHP could fail to complete internal data setup, causing later processing to misinterpret unfinished data as valid. The second vulnerability left PHP holding references to memory from previous requests that had already been freed, allowing subsequent requests to access that invalid memory. Under the right conditions, both could enable remote code execution through Apache/mod_php or application-level PHP-FPM.
PHP assigned both vulnerabilities a CVSS v4.0 score of 9.2, marking them as critical. The fixes shipped in versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12. The potential reach was substantial: as of November 11, W3Techs reported that PHP powered 68.8% of websites whose server-side language could be identified.
How the coordinated release unraveled
The disclosure process deteriorated before Chainguard acted. A developer investigating issues from a 2024 security audit had spotted the out-of-memory problem, but no concrete attack path existed at that time. On August 11, researcher Ryan A. from Calif reported remote code execution from the same underlying flaw, making it PHP Foundation's top priority. OpenAI independently reported the same issue, and Wiz supplied additional attack vectors on August 24, followed by more from OpenAI on September 18.
The coordination broke down in early October. CVE identifiers leaked publicly on October 20 with incorrect release timing and severity information. PHP responded by issuing a warning and accelerating the release to October 21 at 14:00 UTC. Calif then mistakenly published a video demonstrating its proof of concept; though deleted, copies spread. PHP also acknowledged that mitigation guidance shared with critical infrastructure providers may have leaked.
At 07:41 UTC on October 21, Chainguard published its patched images—six hours and 19 minutes before the agreed time. The PHP Foundation objected sharply: "Chainguard unilaterally broke the embargo and published built images containing the fix." The foundation's concern centered on a fundamental problem: a patched binary protects those who install it but also reveals to attackers where to focus their analysis.
A patched binary protects the person who installs it. It can also tell an attacker where to look.
PHP Foundation
Moore's case: attackers already had the advantage
Moore did not dispute that releasing patched binaries can assist attackers. His argument was that the protective value of the embargo had already eroded. When defenders alone know about a vulnerability, confidentiality provides preparation time. Once attackers possess sufficient information to exploit it, withholding available fixes reverses that advantage—attackers can act while users outside the embargo remain exposed.
"If we see exploitation in the wild, it's incumbent upon us to release it," Moore told The New Stack. He attributed Chainguard's decision to reports from other security vendors indicating active exploitation.
According to Moore's account, Chainguard learned of the PHP vulnerability on the Friday before release, pointed its vulnerability-hunting tools at the software, and discovered the flaw or produced a proof of concept within hours. This narrative supports his reasoning: if Chainguard's tools could find the bug quickly, attackers' tools might do the same.
Moore also pushed back against the suggestion that Chainguard acted unilaterally. Other embargo participants contacted CEO Dan Lorenc while PHP maintainers in Europe were offline to discuss the situation, he said, though he declined to name those participants. Moore acknowledged Chainguard made the final call but distinguished between an emergency decision made in consultation and one made in isolation.
What the evidence actually shows
Public records support Chainguard's concern that attackers were already targeting the vulnerabilities. Amazon Threat Intelligence reported a PHP2Shell exploitation attempt against its MadPot honeypot network on October 17, four days before disclosure. It also documented proof-of-concept exploits beginning around 02:00 UTC on October 21, with additional attempts from anonymized infrastructure over the following six hours—some clearly preceding Chainguard's 07:41 release.
Cloudflare reported observing attack traffic targeting its customers before the official release and said its web application firewall protections were enabled by default starting at 03:02 UTC on October 21.
However, Amazon included a critical caveat: "Technical analysis is ongoing, and it's unclear if any of these attempts would have succeeded in real-world systems." The PHP security team examined circulating proofs of concept and "did not identify any functional exploits at the time of release/public disclosure."
These statements need not contradict each other. A honeypot can record an exploitation attempt without proving it achieves remote code execution against a real application. Similarly, the absence of a functional public exploit does not rule out private possession by attackers. On October 26, CISA added both vulnerabilities to its Known Exploited Vulnerabilities catalog, confirming exploitation was known by then—but not establishing when the first successful compromise occurred or whether Chainguard's early release enabled it.
Did the binaries provide new information?
Moore's second line of defense was narrower: attackers were already probing the relevant code, so publishing patched binaries may not have materially helped them. "I don't see any evidence that what we did helped the attackers in ways that they didn't already have," he told The New Stack. He acknowledged that comparing patched and unpatched binaries can identify vulnerable code, including with AI-assisted disassembly, but argued that locating the code differs from building a working exploit.
That distinction holds merit for memory-corruption bugs, where identifying the flaw and reliably turning it into remote code execution represent separate engineering challenges. Yet the defense has boundaries. Evidence that some attackers knew where to look does not establish that all prospective attackers did, nor does it prove the patch contained no useful clues beyond the vulnerability's location. Chainguard's argument depends on how much actionable information was already in circulation and how reliable it was.
The same restraint applies to accusations against Chainguard. Publishing the binaries created an opportunity to study the changes but does not by itself demonstrate that a particular compromise resulted from the early release. The two sides assessed different risks: Chainguard emphasized the danger of withholding a ready patch while hostile activity was visible, while PHP emphasized the danger of distributing new information before the wider ecosystem could update. Both risks were genuine.
The AI factor
A broader concern looms: AI-assisted vulnerability research could compress the window defenders have to respond. The public reports cited do not establish whether pre-release PHP2Shell attempts used AI or whether this was the first incident of its kind.
For Moore, the PHP dispute exemplifies a larger problem: attackers' ability to convert vulnerability information into working exploits may outpace defenders' capacity to build and distribute fixes. He described a scenario in which a Chromium patch becomes public and an attacker develops an exploit faster than compiling the browser takes—a warning about industry direction rather than a claim this exact scenario occurred.
"You literally cannot produce builds fast enough," he said. And building is only the start. Even if Chainguard produces an updated image quickly, customers must incorporate it, rebuild applications, test them, and deploy—a chain of steps that takes time.
Faster does not mean instantaneous, and a supplier's release time is not a customer's remediation time.
Moore's longer-term proposal involves reducing the number of vulnerabilities requiring this race. He discussed work toward ABI-compatible replacements for foundational software components in Rust and formal verification techniques designed to eliminate particular classes of defects. These remain engineering goals awaiting evaluation and do not address what Chainguard should have done on October 21 in the eyes of its critics.
Broader lessons
The incident extends beyond one vendor breaking protocol. It could be read as a story about an agile security company rescuing users from a slow upstream project, but the reality is more complex. PHP's own postmortem identified a deeper problem: a "structural failure" involving both decision-making and insufficient resources.
Finding bugs faster does not automatically provide the engineers to fix them safely. Getting a patch ready at one company does not mean distributions, hosting providers, extensions, and applications are ready to deploy it. Many criticized Chainguard's decision. Jay Gazlay, CISA's acting associate director for vulnerability management, said at an October 28 Red Hat conference: "that type of behavior is dangerous, and we've got to be careful about holding embargoes."
Repair may still be possible. Federal News Network, citing Politico, reported that PHP Foundation Executive Director Elizabeth Barron expressed disappointment with Chainguard but hoped to restore trust and continue working together.
Moore characterized it as "a really tough call" between "two really, really bad choices." He was right. Maintainers and distributors must establish in advance when to lift an embargo early, who decides, and how other distributors will deliver fixes to their users. This needs to happen soon. Otherwise, the next incident will repeat the same pattern: vendors arguing about who broke the embargo while administrators scramble to patch before their servers are compromised.