Summary
act's Artifacts V4 backend does not properly associate artifact operations with the requesting workflow run, allowing reachable clients to access artifacts belonging to other concurrently running jobs. A hardcoded HMAC key and a network-reachable default listener may also enable unauthorized artifact URL generation, reading, replacement, or deletion without credentials.
Affected context
The issue affects act environments that enable the HTTP Artifacts V4 backend through
actions/upload-artifact@v4 or actions/download-artifact@v4. Risk is greatest when the
artifact server listens on a host or network address reachable by untrusted clients,
while control-plane requests are allowed to provide workflow-run identifiers without
verifying ownership and signed artifact URLs rely on a static signing secret.
Exploit preconditions
- Attack surface: The HTTP Artifacts V4 control plane and signed artifact URL endpoints
- Required access: Network reachability to the artifact backend; application authentication is not necessarily required
- Trigger point: Artifact creation, retrieval, listing, finalization, deletion, or signed URL generation for a workflow run
- Why exploitation is possible: The supplied workflow-run backend identifier is parsed but not consistently compared with the requesting task's run ID, while URL authentication uses a hardcoded key and ambiguous message construction
Why this happens
The primary flaw is an authorization failure consistent with CWE-862: Missing Authorization. Artifact control-plane operations trust a caller-supplied workflow-run
identifier without enforcing that it belongs to the requesting task, breaking the
isolation boundary between concurrent jobs.
The design also introduces a related CWE-321: Use of Hard-coded Cryptographic Key
issue. Every build uses the same fixed HMAC key, allowing signatures to be reproduced if
the signing construction is understood. Concatenating fields without explicit length
encoding or delimiters additionally creates ambiguous signed representations, weakening
the integrity model.
Potential impact
- An unauthenticated network client may read artifacts from another running job
- An attacker could overwrite or delete artifacts before the owning job consumes them
- Build outputs may expose secrets, tokens, deployment credentials, or proprietary data
- Artifact replacement could cause downstream jobs to consume attacker-controlled content
- CI/CD integrity and supply-chain trust could be compromised
General mitigation guidance
Mitigation generally focuses on enforcing artifact ownership at every control-plane
operation and reducing the server's network exposure.
Common defensive patterns include:
- Deriving the authorized workflow run from authenticated server-side context rather than trusting caller-supplied identifiers
- Enforcing authorization checks consistently for creation, retrieval, listing, finalization, and deletion
- Binding signed URLs to an unambiguous, canonical representation of all relevant fields
- Using per-deployment, protected signing keys managed through a secure secret mechanism
- Binding signatures to the intended workflow, artifact, operation, and expiration
- Binding the artifact service to loopback or a restricted interface by default
- Applying network policy, firewall rules, and authentication at the service boundary
- Running artifact services with least privilege and isolating jobs from one another
- Treating uploaded artifacts as untrusted and validating them before downstream consumption
How teams detect issues like this
High-level detection can combine source review, runtime monitoring, and deployment policy
enforcement:
- Use static analysis to identify authorization decisions based solely on caller-provided workflow or task identifiers
- Flag commented-out, bypassed, or missing ownership comparisons in artifact handlers
- Search for hardcoded cryptographic keys, static HMAC secrets, and non-canonical string concatenation in signing code
- Review whether artifact endpoints bind requests to authenticated job identity and operation scope
- Monitor unexpected artifact reads, writes, finalizations, and deletions across unrelated workflow runs
- Enforce policies that prohibit artifact services from listening on unrestricted host or network interfaces
- Use automated code and configuration review tools to detect exposed listeners and missing authorization controls

