PHP2Shell: what it means for identity, access and IT governance

Rencore Governance is not affected by PHP2Shell. Rencore Governance does not use PHP as part of the product or any of the subsystems that run it. We continue to monitor emerging threats as they appear.

The incident still matters for every IT and security team. It shows how quickly a critical flaw now moves from disclosure to exploitation, and why knowing what you run decides how quickly you can respond.

What happened

On 21 October 2026, the PHP project released fixes for two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619, known together as PHP2Shell. Both carry a CVSS v4.0 score of 9.2. Both allow remote code execution on servers that use PHP to handle web requests. The fixed versions are PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35.

Four details stand out from the first five days.

• Speed. According to the PHP Foundation, the two CVE identifiers began circulating online around 15:00 UTC on 20 October. That evening, a security firm mistakenly published a video of its proof of concept working against real targets, and PHP moved the release forward by a day. Amazon saw the first exploitation attempts on 17 October, while the flaws were still under embargo, and watchTowr’s honeypots have recorded exploitation since 21 October.
• Exposure. Palo Alto Networks’ Cortex Xpanse telemetry identified more than 22.6 million exposed instances potentially vulnerable to the two CVEs. Across its customers, Qualys found that 71 percent of the PHP installations it could version run branches that are already end of life.
• Precaution. In the Netherlands, MijnOverheid, DigiD, and the website of the Dutch House of Representatives went offline as a precaution on 21 October, NOS reported.
• AI on both sides. Earlier this year the PHP Foundation worked with Alpha-Omega to scan PHP with Anthropic’s Claude Mythos, and it now also uses OpenAI’s Daybreak Red to validate reports and exploits. Calif, OpenAI, and Wiz reported the root cause and further attack vectors between 11 August and 18 September.

If you run PHP: PHP states that no configuration change fully mitigates the vulnerabilities, so updating is the only complete fix. Upgrade to a fixed version, restart every PHP worker so the fix takes effect, and restrict internet access to any system you cannot update yet.

The next target is identity and access

Remote code execution on a web server gives an attacker a foothold. From there, the goal is credentials, identities, and access, because that is what lets an attacker move further into an organization.

That same layer sits under every collaboration estate and every AI rollout. Teams, SharePoint sites, Power Automate flows, Copilot Studio agents, and Claude Enterprise (preview) projects all act through an identity with a set of permissions. Every AI assistant reads at the access those identities already hold. Gartner expects that through 2026, at least 80 percent of unauthorized AI transactions will be caused by internal policy violations, such as oversharing, rather than by malicious attacks.

Non-human identities make this harder. Agents, connectors, and service accounts rarely appear in access reviews, because access reviews were designed for people. When an employee leaves, there is a process. When the project that created an agent ends, there usually is not.

The business case is the cost of not knowing

Every incident like this one is another reminder of why good IT governance matters. Know what you run, who owns it, and what it can reach.

Organizations with an up-to-date inventory and a named owner for every system can answer “are we affected?” in minutes. Without that, the safe choice is to switch services off while teams confirm their exposure. Lost service, staff hours spent searching, and decisions made without facts all carry a cost. Most of it is avoidable.

If you are building the case for governance internally, this incident gives you the evidence. A CISO understands a public exploit video within hours of a leak. A CFO understands services taken offline because nobody could confirm exposure. Measure your own time to answer, and the conversation moves from preference to risk and cost.

Five steps to take this week

1. Time your answer. Pick a platform you rely on. Ask how long it would take your team to confirm whether a critical flaw affects you, without switching anything off. That number is your baseline.
2. Build one inventory. Workspaces, sites, apps, flows, agents, and AI assistants in a single view, each with a named owner. Fix anything without an owner first.
3. Reconcile identities. Compare Microsoft Entra ID with every AI tool and SaaS app you have connected. Look for disabled users who still hold access, guests nobody sponsors, and service accounts with more rights than their job needs.
4. Give non-human identities a lifecycle. Every agent, connector, and service account gets an owner, a purpose, and an expiry date, the same as a guest account.
5. Automate the routine fixes. Decide which issues are safe to remediate automatically and which need approval. Then run policies continuously instead of waiting for the next review.

The first three steps need no new tooling. Each one shortens the time between the next disclosure and a confident answer.

Where Rencore fits

Rencore is a governance platform. We have never called ourselves a data security tool, and we do not scan servers for vulnerable software. Patching stays with your infrastructure and security teams.

Rencore governs the identity and access layer that attackers and AI assistants both reach. We inventory everything people and agents create across Microsoft 365, Power Platform, Copilot, and AI agents, with Claude Enterprise and ChatGPT Enterprise in preview. We show who and what can reach which data, track ownership, automate lifecycle, and run policies continuously. One Microsoft Entra ID identity joins it all, so access, ownership, and spend read the same way whether the actor is a person, a flow, or an AI assistant. For Claude Enterprise, Rencore flags users disabled in Entra ID who still hold Claude access.

Governance is what lets an enterprise say yes to AI, not the brake on it.

See how quickly your organization could answer “are we affected?” Talk to an expert, or start a free 14-day trial.

Sources

1. The PHP Foundation, Today’s PHP security release, and how we’re strengthening PHP security, 21 October 2026.
2. watchTowr, PHP2Shell FAQ: CVE-2026-109618 and CVE-2026-109619.
3. SecurityWeek, PHP2Shell exploitation started days before patches, citing Amazon.
4. Palo Alto Networks Unit 42, Threat Brief: Critical vulnerabilities in PHP: CVE-2026-109618, CVE-2026-109619.
5. Qualys, Threat Research Unit analysis of PHP2Shell exposure, October 2026, as reported by SecurityBrief.
6. NOS, Websites als MijnOverheid en DigiD offline vanwege lek, 21 October 2026.
7. Gartner, Market Guide for AI Trust, Risk and Security Management, 18 February 2025.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organisation and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.

Last updated 26 October 2026