Critical PHP flaws are being attacked—what must website operators do now?
PHP has patched two critical flaws that can allow unauthenticated remote code execution. Cloudflare says it has already seen attack traffic: operators should update to a fixed PHP release, even if a firewall rule is in place.
The short version
- What happened
- PHP says CVE-2026-109618 and CVE-2026-109619 can permit unauthenticated remote code execution. Fixed releases are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35.
- Who it affects
- People using affected websites could face disruption, data exposure or downstream compromise. Small organisations should ask their host which PHP build is running and whether the fixed release or a verified backport is installed.
- What remains unproven
- Public sources do not provide a reliable denominator for vulnerable or compromised servers.
Update PHP now; a firewall is only a temporary layer
The PHP project has released fixes for two critical vulnerabilities that can let an unauthenticated attacker execute code on a server. The fixed branches are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. PHP says partial workarounds exist, but updating is the only complete protection. PHP 8.1 and earlier no longer receive normal project updates; operators should obtain a supported distribution backport or move to a supported release.[1][2][3]
Who it affects
People using affected websites could face disruption, data exposure or downstream compromise. Small organisations should ask their host which PHP build is running and whether the fixed release or a verified backport is installed.[1][2][3][4]
Keep in mind
Public sources do not provide a reliable denominator for vulnerable or compromised servers.
Cloudflare's observed traffic covers its network and cannot establish global attack volume.
Exposure depends on PHP version, configuration and reachable request-handling paths.
OpenAI and Wiz have commercial interests; PHP is the authoritative patch record.
Multi-source analysis · 4 linked sources. Original source date: 21 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 3 min The evidence, methods, implications and limitations
Update PHP now; a firewall is only a temporary layer
The PHP project has released fixes for two critical vulnerabilities that can let an unauthenticated attacker execute code on a server. The fixed branches are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. PHP says partial workarounds exist, but updating is the only complete protection. PHP 8.1 and earlier no longer receive normal project updates; operators should obtain a supported distribution backport or move to a supported release.[1][2][3]
Attack traffic makes this an active incident
Cloudflare says it observed exploitation attempts and mutated payloads, and deployed managed protections about nine hours before public disclosure. The flaws affect PHP's core request handling and are not limited to one application framework. Exposure still depends on configuration and request paths, so PHP's large footprint should not be read as a count of vulnerable servers. Defenders should review internet-facing hosts, containers, appliances and managed services separately.[2][3]
The distinction that matters
The defensive sequence
Reduce immediate exposure, then remove the underlying vulnerability.
Find
Inventory PHP versions and exposed paths
Identify affected services
Contain
Apply WAF and temporary mitigations
Reduce exploit reach
Fix
Install a patched release or backport
Remove the known flaws
AI accelerated discovery, not the evidence standard
OpenAI says its research system identified one flaw while Wiz independently found the vulnerability family, followed by human verification and coordinated disclosure. That is consequential evidence that AI can lower the time and expertise needed to explore complex code. It is not proof that an autonomous system safely completed the investigation. The assessment should be updated when patch coverage, exploitation volume and confirmed compromise data become available.[3][4]
Evidence check
Public sources do not provide a reliable denominator for vulnerable or compromised servers.
Cloudflare's observed traffic covers its network and cannot establish global attack volume.
Exposure depends on PHP version, configuration and reachable request-handling paths.
OpenAI and Wiz have commercial interests; PHP is the authoritative patch record.
Reporting basis: Multi-source analysis. Original source date: 21 October 2026.
Continue the story
Read next
Continue exploring cybersecurity: compare the finding and its evidence limits.
What did the agent-security recreation actually prove?
A new unreviewed paper reproduced a published JFrog privilege-escalation flaw inside deliberately vulnerable containers and connects it to the 2026 OpenAI–Hugging Face incident. The recreation demonstrates an infrastructure failure mode; it cannot establish what undisclosed training data caused the agents' behaviour.
Read the next reportExplore the full source and evidence record
What this means for people
- People using affected websites could face disruption, data exposure or downstream compromise. Small organisations should ask their host which PHP build is running and whether the fixed release or a verified backport is installed.
Global context
PHP underpins websites and managed applications worldwide, including services maintained by small organisations that may not control their runtime directly. Unsupported branches and uneven vendor backports make supplier confirmation important.
What the evidence does not yet show
- Public sources do not provide a reliable denominator for vulnerable or compromised servers.
- Cloudflare's observed traffic covers its network and cannot establish global attack volume.
- Exposure depends on PHP version, configuration and reachable request-handling paths.
- OpenAI and Wiz have commercial interests; PHP is the authoritative patch record.
What to watch next
- Confirmed compromise indicators
- Distribution and hosting-provider backports
- Patch uptake and successful exploitation
- Postmortems separating AI leads from human verification
Living evidence record
Impact record IAI-1Y13T75
Evidence stage
Announced
Confidence
Supported
Reporting basis
Multi-source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
21 October 2026
Source trail
4 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Evidence trail
Sources used for this report
Links checked 21 October 2026
Research question, topics and regions
Topics: PHP · Cybersecurity · Remote code execution · Web infrastructure · AI-assisted vulnerability research
Regions: Global
This report is labelled multi-source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.