An urgent PHP advisory arrives while your team is already delivering customer work. You need to identify affected servers, update them and make sure the customer services still work. OpalSpan lets you give that device work to AI agents: Mech in the browser or your own assistant through Agent CLI.
Consider an MSP responsible for an illustrative fleet of 50 servers. Some host WordPress sites, some run customer applications, and others contain PHP that nobody associates with a public website. OpalSpan provides the authorized connection and agent workspace for investigating those machines and carrying out their individual fixes. The useful outcome is a fleet whose actual status is known, with completed updates and named exceptions you can act on.
CVE-2026-109618 and CVE-2026-109619, discussed as php2shell, are the immediate reason to do this. The 21 October PHP advisory calls for urgent security updates. Finland's government ICT provider Valtori described precautionary restrictions on public-internet services, lifted service by service after corrective work and checks. Its statement explicitly distinguishes those precautions from a breach.
An agent-ready fleet gives a service provider a practical way to work through that pressure. Here is the workflow to put in place.
Give each customer fleet a clear job
Start with the servers your team is authorized to manage in the appropriate OpalSpan workspace. Preserve customer boundaries and select the actual devices for the job. Connect missing servers through the approved enrollment or attended-access process before expecting agent work on them.
Tell Mech what completion looks like:
Check the selected customer servers for the PHP security issues CVE-2026-109618 and CVE-2026-109619. Identify all PHP web-serving runtimes and the package, image or hosting vendor responsible for them. Classify each target using current vendor fix evidence. For approved maintenance updates, prepare recovery, patch a representative canary and verify its customer service before continuing through the group. Keep redundant application capacity available. Report offline, inaccessible, unsupported and compatibility-blocked targets separately, with the next action for each. Do not cross customer scope, perform an unapproved branch migration or reboot the whole group.
The 50-server example illustrates an operating problem, not measured OpalSpan throughput. Choose a workload and degree of parallelism that your available agents, model capacity, application redundancy and change policy support. The instructions define the agent's job; the platform still applies the user's actual device permissions.
This is valuable even before the first package changes: agents can perform the repetitive environment inspection while your team resolves the exceptions requiring customer knowledge or a commercial decision.
Have agents establish exposure on the machines
A spreadsheet saying “PHP 8” is not enough to select a security update. Ask the agent working on each server to trace the application to its actual PHP handler and record the observation time.
The investigation should establish:
- The website or application, its serving PHP version and whether it uses FPM, mod_php, CGI, a container or another supported arrangement.
- The exact distribution package, repository, container image or hosting-provider build, with the vendor's fix status.
- Additional pools, workers, virtual hosts or embedded installations that may need separate attention.
- The supported update path, service owner, approved maintenance window and recovery route.
- Any access or evidence gap that prevents a conclusion.
OpalSpan's agents can execute the relevant inspection commands, inspect configuration and collect the results through their authorized access. They use established vendor and operating-system tools; the PHP advisory supplies the vulnerability facts.
For this emergency, upstream's first fixed releases are 8.2.35, 8.3.36, 8.4.27 and 8.5.12. Distribution backports may use different version strings. Treat the vendor advisory and installed build together, as described by PHP and CERT-EU. A host without enough evidence belongs in the unresolved group, not the safe group.
Turn the findings into different repair paths
The advantage of giving agents a goal is that the next action can follow the machine's actual condition. These fictional findings illustrate the decisions:
| Target | Agent finding | Useful next action |
|---|---|---|
| web-01 | Supported PHP branch; fixed vendor package available | Back up, apply the approved maintenance update, replace affected workers and test the site |
| web-02 | Older-looking package with a verified vendor backport | Record the vendor evidence and verify the running service uses that fixed build |
| shop-03 | Legacy WordPress plugin blocks a supported PHP migration | Prepare staging and the compatibility work before proposing a production change |
| app-04 | PHP comes from an application container image | Update through the approved image/deployment owner and test the replacement instance |
| web-05 | Server is offline or access is unavailable | Keep the unresolved exposure visible and identify who can restore access |
For an agency, the third row is often where the real work begins. Mech can investigate and carry out a WordPress compatibility upgrade, including plugin updates and application checks. An abandoned plugin, new licence or customer-facing redesign still needs its proper decision owner.
This turns a broad alert into deliverable customer work. Your team can see which repairs agents can complete within the existing authority and which need a different plan.
Roll out through the fleet while keeping services available
Group servers by the change they actually need: package source, PHP branch, application dependency and operating role. A canary on one stack does not establish compatibility for every other stack.
For a redundant web service, an approved workflow may drain one instance from traffic, have an agent update and check it, return it to service, then continue with the next instance. Mech can perform the server-side work and use existing deployment or load-balancer tooling when that access is available and authorized. It should identify a missing control path rather than imply that a hosting-panel or cloud integration exists automatically.
For a single-server customer site, agree on the required maintenance window. A brief protective interruption can be the right operational choice. OpalSpan's contribution is agent capacity to prepare the update, execute it and verify recovery, reducing avoidable delay and repeated manual work.
Have the agent distinguish a PHP-FPM or web-service restart from an operating-system reboot. Restart or replace the processes required by the vendor's update procedure; confirm they no longer use vulnerable code. A reboot needs an approved window and recovery access. Confirm that the device reconnects and that current authority permits continued work; boot-scoped attended approval does not automatically survive a reboot.
A failed canary should stop expansion of that change. Preserve the observed failure, restore service through the agreed recovery path and let the agent investigate the specific incompatibility. If recovery would return to vulnerable PHP, retain the necessary exposure restriction while resolving the update.
Ask for service results, not a fleet-wide guess
After each change, have the agent verify the serving runtime and exercise the customer service's agreed checks. A process running, a package installed and a checkout working are different observations. Preserve those distinctions in the conversation and any relevant customer work record.
At the end, ask Mech for a concise result organized around action:
- Servers updated and customer-service checks passed.
- Servers already fixed, with the supporting vendor/build evidence.
- Targets requiring compatibility work, hosting-provider action or a different change window.
- Offline or inaccessible targets whose exposure is still unresolved.
Keep a time on the assessment. A newly deployed container, restored backup or newly enrolled server can change tomorrow's position. A completed patch also does not establish that earlier compromise was absent; suspicious evidence needs its own incident-response work.
Put agent capacity into your customer service
The commercial benefit is straightforward: your team can give agents more of the investigation, maintenance and verification work customers are paying you to deliver. OpalSpan keeps that work connected to the authorized devices and available conversations. You retain the decisions that need customer context while agents carry out the supported technical work.
Try OpalSpan on a small approved server group. Give Mech the inventory-and-remediation task, inspect the results, and use the current plan and credit options to expand the workload that fits your operation. Your own assistant can use Agent CLI as another way to put agent capacity on the same authorized devices.
Once the immediate fleet work is complete, give Mech a recurring PHP maintenance responsibility so the next advisory starts with a prepared fleet.
Sources
Security facts checked 22 October 2026; package availability and exposure require a current per-target check.
- PHP: critical remote-code-execution vulnerabilities and emergency updates.
- Official CVE record: CVE-2026-109618 and CVE-2026-109619.
- PHP GitHub advisory: RCE via OOM corruption and RCE via stale global state.
- CERT-EU: Critical Vulnerabilities in PHP.
- Valtori: PHP update progress and service-by-service restoration, updated 22 October; the separately reported publishing-platform maintenance was unrelated to this PHP event.