CVE-2026-46477: Flowise: Dataset create+update mass-assignment allows cross-workspace dataset takeover

Synopsis

A mass-assignment vulnerability in Flowise, the drag-and-drop LLM flow builder, allows an authenticated attacker to reassign or take over datasets belonging to other workspaces by injecting unexpected fields into dataset create or update requests. The attack is reachable over the network and requires only a low-privilege account within the application. Successful exploitation gives the attacker full read, write, and denial-of-service capability over victim workspace datasets. A patched-image rebuild at version 3.1.2 is available on HarborGuard for environments running an affected version.

HarborGuard Coverage

Exploit Conditions

• Network reachabilityRequired

  The attacker must reach the Flowise API over the network; the service must be exposed to the attacker's network segment.

• AuthenticationRequired

  A low-privilege Flowise account is sufficient; no administrative or elevated permissions are needed.

• Victim interactionNot required

  No user interaction is required; the attacker acts entirely through their own crafted requests.

• Attack complexityDetail

  Base exploit reliability is high, though AT:P indicates a specific deployment condition (such as multi-workspace configuration) must be present for the cross-workspace takeover to succeed.

Blast Radius

• Attacker reads dataset contents belonging to other workspaces, including any training data, prompts, or user-supplied records stored there.
• Attacker overwrites or reassigns dataset records in victim workspaces, corrupting LLM pipeline inputs or substituting malicious data.
• Attacker can render datasets unavailable to their legitimate workspace owners, disrupting active LLM flows that depend on them.