Skip to content

Image: seclists.org · rights & removal

Executive Summary

A Fastly Next-Gen WAF agent running as a Kubernetes sidecar has a design choice that results in a fail-open security posture when the agent has not yet loaded its ruleset. When the agent starts without configuration, it passes all traffic uninspected because it cannot connect to the necessary RPC socket for rule loading. This state persists until configuration is successfully downloaded; during this time, every request is processed with no WAF mitigation applied. The system's readiness status in Kubernetes does not reflect whether rules have loaded or if the agent is operating securely. Mitigation involves configuring a startup probe, which causes the pod to remain unavailable (503) until the agent can successfully load its configuration and establish necessary connections. Vendor position maintains that the fail-open behavior is an intentional design choice prioritizing application availability during unavailability, not a vulnerability in itself.

Facts Only

* The Fastly Next-Gen WAF agent operates as a Kubernetes sidecar deployment using the NGINX module and a `sigsci-agent` sidecar sharing a Unix socket.
* When the agent has not loaded a ruleset, the WAF passes traffic uninspected while reporting readiness.
* Before version 4.52.0, no mitigation existed; versions 4.52.0 and later are affected unless the startup probe is configured.
* The vulnerability relates to failing open (CWE-636), insecure initialization of a resource, and incorrect behavior order.
* The mechanism involves the agent failing to connect to its RPC socket during initialization, leading it to pass requests through without rules.
* Lab testing confirmed that in default deployments, every request received a WAFResponse 200 with the header "X-SigSci-Agent: disabled" until configuration loaded.
* Startup failures (e.g., connection drops) caused a window where up to 935 requests passed during the no-socket phase when using slow failure modes.
* Mitigation requires configuring the agent startup probe with `SIGSCISTARTUPPROBELISTENER` to enforce a delay before processing traffic.
* Vendor position is that fail-open is a deliberate design choice prioritizing availability, not a vulnerability.

Full Take

The narrative centers on a fundamental tension between security posture and operational availability within a dynamic container orchestration environment. The core implication is that a system designed for resilience defaults to maximum permissiveness when initialization fails, shifting the burden of security correctness from runtime verification to deployment configuration. This structure mirrors historical patterns where operational continuity (availability) often supersedes perfect security enforcement during transitional states.
The pattern of "fail-open" suggests a design philosophy rooted in preventing service denial during infrastructure instability, yet this choice demonstrably allows uninspected traffic flow. The vendor's justification—prioritizing availability—frames the vulnerability not as an oversight but as a trade-off accepted within a specific architectural constraint. The lack of documentation regarding what behavior results from an agent answering with no ruleset loaded highlights a gap in transparency that exacerbates reliance on default settings without explicit operator awareness.
The consequence for human agency lies in the necessity to engineer security behavior explicitly, rather than relying on opaque defaults. When operators are not informed about failure states (like the agent remaining "disabled" but passing traffic), they cannot effectively measure or enforce the desired security state. The pattern detected is a systemic shift where complex, distributed systems embed risk by prioritizing abstract availability over explicit, immediate security verification at critical initialization points.
Bridge questions: How should system architects weigh the absolute necessity of operational uptime against the principle of least privilege during agent initialization? What mechanisms can be built into control planes to enforce verifiable security states rather than relying solely on application-level behavior? How does this incident change the expectation for vendor documentation regarding default security behaviors in Kubernetes deployments?

From the original · Full Disclosure mailing list

Full Disclosure mailing list archives Fastly Next-Gen WAF (Signal Sciences) agent fails open before rules load in Kubernetes sidecar deployments (vendor disputes; CVE pending) From: Zac Lohrenz Date: Tue, 17 Nov 2026 17:29:12 -0600 Fastly Next-Gen WAF (Signal Sciences) agent: fail-open before rules load in Kubernetes sidecar deployments Product: Fastly Next-Gen WAF agent (sigsci-agent) with the…
Read the full story at seclists.org

Sentinel unavailable

The automated check of the source article's wording did not complete, so no result is shown.

This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.

Fastly Next-Gen WAF (Signal Sciences) agent fails open before rules load in Kubernetes sidecar deployments (vendor disputes; CVE pending) | Huntaegis