Cloudflare published research on August 19, 2026 demonstrating a Spectre-class side-channel attack against Cloudflare Workers capable of stealing JWT authentication tokens at 12 bits per second—360 times faster than prior techniques. The attack exploits speculative execution in V8 JavaScript engine and uses WebSocket keep-alive messages within Durable Objects to evade behavioral detection. Cloudflare confirmed no evidence of real-world exploitation and stated that three layers of mitigation are already deployed across production infrastructure.
Security researchers disclosed 'GhostJacking,' a novel attack class exploiting Cloudflare's WAF logs to manipulate AI agents. Attackers inject malicious instructions into blocked-request logs that agents misinterpret as operational guidance. Testing showed 9 of 10 successful attempts to convince agents to alter DNS settings under Cloudflare-recommended configurations. The attack leverages legitimate agent permissions and approved tools, bypassing traditional endpoint and WAF detection.
Cloudflare's repositories were among 300+ high-impact open-source projects identified as vulnerable to Cordyceps, a CI/CD attack pattern disclosed in June 2026 by Novee Security. The weakness allows attackers with a free GitHub account to exploit malicious pull requests targeting pull_request_target and workflow_run triggers, potentially achieving code execution and access to repository secrets. Affected projects include those by Microsoft, Google, Apache, and the Python Software Foundation; all pipelines passed security checks despite exposure.
An HTTP/2 vulnerability dubbed 'HTTP/2 Bomb' affects Cloudflare and other server platforms including NGINX, Apache, IIS, and Envoy. The flaw exploits HPACK header compression and HTTP/2 flow control mechanisms, allowing a single client to consume up to 32GB of server memory in approximately 20 seconds, causing denial-of-service conditions and server outages.
Cloudflare disclosed results from Project Glasswing, an internal exercise in which its security team ran Anthropic's unreleased Claude Mythos Preview model against more than fifty of its production repositories, including runtimes, edge data paths and protocol stacks. The reported findings include 595 crashes and ten documented working control-flow hijacks, with the model chaining low-severity primitives into functional exploit chains and surfacing memory-safety and command-injection issues in Cloudflare code. Anthropic rated Mythos at or near its ASL-3 cyber-capability threshold and is keeping the model under restricted release. Cloudflare's CISO framed the exercise as evidence that AI-assisted vulnerability discovery is now operationally relevant against critical infrastructure code, and warned that the same capabilities will be available to attackers. No active exploitation has been reported, and Cloudflare has not published specific CVEs tied to the findings.
Cloudflare published a response detailing how it assessed and mitigated the 'Copy Fail' Linux kernel local privilege escalation vulnerability (CVE-2026-31431), publicly disclosed on April 29, 2026. Cloudflare's Security and Engineering teams reviewed the exploit technique, evaluated exposure across its infrastructure, and validated that existing behavioral detections could identify the exploit pattern. The post describes mitigations applied to protect Cloudflare's fleet from local privilege escalation via the kernel flaw. No active exploitation against Cloudflare was reported, and the issue is a kernel-level LPE rather than a remote-facing service vulnerability. Other items in the source set concern unrelated vendors (Ivanti EPMM, Palo Alto PAN-OS) and are not relevant to Cloudflare.