Nine hours before the PHP2Shell flaws were disclosed in October 2026, Cloudflare said it enabled default web application firewall protections for its customers. AWS had identified an attempted exploit four days before disclosure.
Key takeaways
- AWS analysts identified an attempted PHP2Shell exploit on October 17, four days before public disclosure and patching.
- Cloudflare said its PHP2Shell WAF mitigations were enabled by default for free and paid customers about nine hours before disclosure.
- The PHP Foundation limited its patches to PHP 8.2 and newer, leaving older deployments reliant on third-party vendors.
- Cloudflare said a November bug caused the loss of 55% of logs pushed to customers over 3.5 hours.
- Cloudflare acquired Deno in October 2026; Deno had developed an open-source alternative to Workers.
Cloudflare is becoming an edge intervention layer: a distributed point that can filter traffic, enforce policy and increasingly host application work before requests reach origin infrastructure. That compresses the time between recognizing a threat and protecting customers. It also puts more security and execution decisions—and more consequences of a bad decision—inside one network.
Default protection replaced a customer decision in the PHP2Shell response
In 2019, Cloudflare offered Bot Fight Mode as an opt-in feature for accounts. Customers chose whether to turn on its computational challenges. In 2021, Cloudflare added Email Routing and an email-security DNS wizard intended to help prevent phishing and spoofing. The company was extending its protective reach, but the bot-control example still depended on customer election.
Cloudflare said its PHP2Shell WAF rules took a different route in October 2026. It enabled mitigations by default for free and paid customers about nine hours before disclosure. That does not mean every Cloudflare security feature became mandatory. It means this particular response did not wait for each customer to find the alert, evaluate a rule and switch it on.
The distinction matters operationally. An opt-in control depends on the customer’s response time. A default rule substitutes Cloudflare’s response time—and Cloudflare’s judgment about what legitimate traffic the rule might affect.
Attackers were already working before the disclosure clock started
AWS analysts identified an attempted PHP2Shell exploit on October 17, four days before public disclosure and patching. Cloudflare CEO Matthew Prince then reported thousands of attempts on Cloudflare’s network in the ten hours before disclosure, tens of thousands within 45 minutes afterward, hundreds of thousands within two hours and millions later.
Those are attempts observed on Cloudflare’s network, not a count of successful compromises or attackers. Nor do they measure how many requests the new rules stopped. They do establish the pace an origin operator faced. Patching remained the durable fix; Cloudflare called its WAF rules an evolving stopgap and urged customers to update PHP.
Some origins could not take even that next step directly. The PHP Foundation limited its patches to PHP 8.2 and newer, leaving older deployments dependent on third-party vendors. An edge rule can narrow exposure while an operator waits. It cannot remove the vulnerable code, and October 17 shows that even a pre-disclosure rule can follow the first adversary.
Workers gives the traffic boundary more to govern
Cloudflare launched Workers AI in 2023 with pay-as-you-go access to partners’ GPUs for customers running models. In 2024, it introduced Firewall for AI as a layer in front of LLMs to identify abuse before requests reached a model. The two products occupy different positions: one helps run model workloads; the other inspects traffic headed toward them.
Cloudflare kept extending the developer side of that boundary. Kitesurf, introduced in August 2026, is a cloud-hosted browser for AI agents built on Workers. In October, Cloudflare acquired Deno, whose company had developed an open-source alternative to Workers. Neither move proves that every protected application runs on Cloudflare, or that its partner GPUs sit at every edge location. Both give developers more reasons to build where Cloudflare already sees requests.
That position is distinct from ownership of the financed, place-bound capacity behind the AI cloud capacity market. Cloudflare’s partner-GPU arrangement does not establish who owns that capacity. Its documented advantage here is the opportunity to make a policy decision before a request proceeds to an application or model.
Bundle growth does not identify the winning product
Cloudflare reported $1.67 billion in fiscal 2024 revenue, up 29% year over year. It had 173 customers spending more than $1 million annually, up 47%. By Q3 2026, quarterly revenue reached $782.5 million, up 39%, and Cloudflare raised its full-year revenue guidance to about $2.96 billion. It also reported a $34.1 million net loss for the quarter.
Cloudflare appeared in 12 articles, or 0.13 per day, in the first quarter of 2026 and 25, or 0.27 per day, in the second, before PHP2Shell. That is a change in editorial attention, not a measure of security sales. Cloudflare’s reported results likewise do not break out delivery, security and edge-compute revenue. Larger customers and faster growth support demand for the bundle; they do not show which component captures its value or whether CDN economics have been left behind.
Cloudflare also faces companies with substantial security businesses. Zscaler reported Q3 revenue of $850.5 million, up 25%, while Palo Alto Networks reported Q1 revenue of $3.45 billion, up 39%. Their quarters and products are not like-for-like measures of Cloudflare’s edge strategy. Fastly’s documented acquisition of Glitch and later closure of its hosting service is no evidence of a comparable default-security response, either. Cloudflare’s path should not be mistaken for a proven sector-wide transition.
A shared defense needs shared-failure discipline
Cloudflare attributed a major November 2025 outage to a database-permissions change, not malicious activity. A separate outage in December lasted 25 minutes, following the prior month’s hours-long disruption. Cloudflare also said a November bug caused it to lose 55% of logs pushed to customers over 3.5 hours. For customers relying on its network, availability and visibility are parts of the protection, not ancillary services.
Those incidents do not show that the PHP2Shell rules failed. They show what an operator must assess before allowing a provider to make default decisions at scale: how it checks a rule for false positives, limits a faulty change and reverses it under load. The available account of the PHP2Shell rollout does not establish those mechanics.
Frequently asked questions
How did Cloudflare test the default PHP2Shell rules for false positives before enabling them broadly?
The available account does not establish the testing, canarying, or false-positive-validation process used for the rollout. It also does not describe what traffic exceptions, if any, the rules created.
What rollback or containment process was available if the PHP2Shell rules disrupted legitimate customer traffic?
The piece does not document a rollback mechanism, change-limiting procedure, or recovery time for the rules. Those mechanics remain important because Cloudflare made the decision centrally for free and paid customers.
How many Cloudflare customers were running vulnerable PHP versions when PHP2Shell was disclosed?
That number is not provided. The reported exploit-attempt volumes measure requests observed on Cloudflare’s network, not the number of vulnerable customer origins or successful compromises.
Did Cloudflare’s acquisition of Deno change how Workers operates or how Deno customers are migrated?
The piece confirms the October 2026 acquisition but does not document an integration plan, product migration, or changes to Workers’ technical architecture. It only establishes that Cloudflare brought in a company that had built an open-source Workers alternative.
PHP2Shell exploit-response clock
- October 17, 2026 — AWS analysts identified an attempted PHP2Shell exploit, four days before public disclosure and patching.
- About 9 hours before public disclosure — Cloudflare enabled default WAF mitigations for free and paid customers.
- 10 hours before disclosure — Cloudflare CEO Matthew Prince reported thousands of exploit attempts on Cloudflare’s network.
- 45 minutes after disclosure — Reported attempts had risen to tens of thousands.
- 2 hours after disclosure — Reported attempts had risen to hundreds of thousands.
- Later — Cloudflare reported that attempts reached millions.
Nine hours was a head start on public disclosure, not on the first attacker. It was also nine hours during which Cloudflare’s rule, rather than each customer’s response time, governed exposure. The shorter clock is valuable. The number matters because one network now has to keep that clock—and its own changes—under control.