Chainstack described an intrusion that started from an exposed development Grafana instance with default credentials. Its data-source proxy could reach an internal Kubernetes MCP server without independent authentication, and commands there ran with cluster-administrator authority. The investigation reports changes to ConfigMaps and automation workflows,…
Inspect the ClaimsCase · DiggingBeagle record
Chainstack compromise crossed an unauthenticated Kubernetes MCP control plane
Chainstack described an intrusion that started from an exposed development Grafana instance with default credentials. Its data-source proxy could reach an internal Kubernetes MCP server without independent authentication, and commands there ran with cluster-administrator authority. The public incident account does not prove every action was model-chosen, that every accessible secret was exfiltrated or that the ARTEX tool lineage identifies the same operator a
The investigation reports changes to ConfigMaps and automation workflows, stolen infrastructure credentials, accesses to production data and persistence attempts across Kubernetes resources. Some production database values were returned through command output, while individual MCP call records were not preserved.
Assessment profile
Assessment method- Setting
- production
- Exploitation
- confirmed in the wild
- Evidence
- Grade B
- Remediation
- mitigation available
- AI role
- WITH AI
Basis and provenance
production setting with confirmed_in_the_wild status, based on the identified primary disclosure and its stated test or incident limits. The new Claim distinguishes source observations from hypothetical downstream effects. Impact band is unassigned because the scoped evidence does not justify a comparable quantitative score; no unobserved customer harm or loss is imputed. Remediation state: mitigation_available.
Assessed Nov 13, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-ENR243-007-DETAIL · CLM-R242-CHAINSTACK-GRAFANA-MCP-PIVOT-LIMITS · CLM-R242-CHAINSTACK-GRAFANA-MCP-PIVOT-MECHANISM · Chainstack November 2026 security incident reconstruction
At a glance
A supposedly internal tool server became an authority escalation route after the frontend fell: network reachability and proxy forwarding substituted for authentication at the effectful backend.
Read the implicationsThe public incident account does not prove every action was model-chosen, that every accessible secret was exfiltrated or that the ARTEX tool lineage identifies the same operator as the Barracuda intrusion.
Limits and uncertaintyFull account
Chainstack described an intrusion that started from an exposed development Grafana instance with default credentials. Its data-source proxy could reach an internal Kubernetes MCP server without independent authentication, and commands there ran with cluster-administrator authority.
The investigation reports changes to ConfigMaps and automation workflows, stolen infrastructure credentials, accesses to production data and persistence attempts across Kubernetes resources. Some production database values were returned through command output, while individual MCP call records were not preserved. A supposedly internal tool server became an authority escalation route after the frontend fell: network reachability and proxy forwarding substituted for authentication at the effectful backend.
The public incident account does not prove every action was model-chosen, that every accessible secret was exfiltrated or that the ARTEX tool lineage identifies the same operator as the Barracuda intrusion. Bind MCP tool execution to an authenticated principal with restricted cluster roles, and reconstruct persistent credential use and database effect logs.
Evidence reconciliation and effect boundary. A development dashboard became an authentication bypass into a more privileged tool surface because its datasource proxy could reach the internal MCP service. The attacker then reused machine credentials and cloud permissions in later stages; the existing public account separates reachable secrets, observed reads and confirmed transfers. The infrastructure evidence supports a privilege-boundary failure, but missing individual MCP call logs limit conclusions about exactly which AI agent chose each command.
Workflow-history bypass and production pivot. Chainstack places the exposed developer Grafana/default credential access on October 26 and describes its datasource proxy reaching an unauthenticated cluster-admin Kubernetes MCP service. After moving into other credentials, attackers modified an automation workflow directly in its database to run a shell command, then restored the workflow, thereby avoiding the ordinary application modification trail. Later access included production secret-manager reads, cloud credentials and an unauthorized CI runner. A Grafana compromise, an MCP authority exposure and a database-level workflow rewrite are separate boundaries; individual agent decisions and actual MCP tool call logs were not reconstructed.
Mechanism and trust boundary
- 01
The investigation reports changes to ConfigMaps and automation workflows, stolen infrastructure credentials, accesses to production data and persistence attempts across Kubernetes resources. Some production database values were returned through command output, while individual MCP call records were not preserved.
- 02
A supposedly internal tool server became an authority escalation route after the frontend fell: network reachability and proxy forwarding substituted for authentication at the effectful backend.
Claims & evidence
CLM-ENR243-007-DETAILChainstack describes a developer Grafana deployment with default credentials and a datasource proxy that reached an internal Kubernetes MCP service with cluster-admin authority; that path preceded access to production secrets and data.supported
Basis: reported finding
- supportsChainstack November 2026 security incident reconstructionprimary disclosure
Chainstack November incident report, initial entry and Grafana datasource proxy to MCP/Kubernetes account chronology
CLM-NOV15-CHAINSTACK-DATABASE-EDITChainstack reports database-level modification and restoration of an automation workflow during an intrusion that also traversed a default-credential Grafana and a cluster-admin Kubernetes MCP server.supported
Basis: reported finding
- supportsChainstack November 2026 security incident reconstructionprimary disclosure
Chainstack incident report, October 26-31 chronology, Grafana datasource proxy, DB-edited workflow and production secret-manager access
CLM-R242-CHAINSTACK-GRAFANA-MCP-PIVOT-LIMITSThe public incident account does not prove every action was model-chosen, that every accessible secret was exfiltrated or that the ARTEX tool lineage identifies the same operator as the Barracuda intrusion.supported
Basis: reported finding
- supportsChainstack November 2026 security incident reconstructionprimary disclosure
Incident report, Grafana entry, internal MCP access, credential use and Kubernetes changes
CLM-R242-CHAINSTACK-GRAFANA-MCP-PIVOT-MECHANISMChainstack described an intrusion that started from an exposed development Grafana instance with default credentials. Its data-source proxy could reach an internal Kubernetes MCP server without independent authentication, and commands there ran with cluster-administrator authority.supported
Basis: reported finding
- supportsChainstack November 2026 security incident reconstructionprimary disclosure
Incident report, Grafana entry, internal MCP access, credential use and Kubernetes changes
Implications
A supposedly internal tool server became an authority escalation route after the frontend fell: network reachability and proxy forwarding substituted for authentication at the effectful backend.
Controls and mitigations
- Bind MCP tool execution to an authenticated principal with restricted cluster roles, and reconstruct persistent credential use and database effect logs.
- Retain immutable database effect logs and correlate workflow storage mutations against application audit events, including out-of-band writers.
Unknowns and contradictions
- The public incident account does not prove every action was model-chosen, that every accessible secret was exfiltrated or that the ARTEX tool lineage identifies the same operator as the Barracuda intrusion.
Sources and citation
Material revision history
- Nov 15, 2026 · Canonical change recorded · new in release · revision 250