Arcade — agentic threat model
Arcade acts as a high-privilege integration layer, delegating user identity to LLMs for tool execution across sensitive platforms like GitHub and Gmail. Its primary risk lies in prompt injection translating directly into unauthorized API actions, mitigated partially by its dedicated Agent Auth and VPC deployment capabilities.
OWASP AIVSS score rationale
| Autonomy of Action | 0.80 | |
| Goal-Driven Planning | 0.70 | |
| Self-Modification | 0.20 | |
| Dynamic Tool Use | 0.90 | |
| Persistent Memory | 0.50 | |
| Contextual Awareness | 0.80 | |
| Dynamic Identity | 0.90 | |
| Multi-Agent Interactions | 0.30 | |
| Non-Determinism | 0.70 | |
| Opacity & Reflexivity | 0.50 |
Scored with the canonical OWASP AIVSS formula (AIVSS calculator reference); agentic risk factors estimated from the agent’s described capabilities.
MAESTRO 7-layer threat model
Per-layer threats for this agent. Layers tagged “not certain from listing” are general, caveated commentary where the public description didn’t pin that layer.
Not certain from the listing — Arcade is model-agnostic, focusing on LLM-Tool interactions. Threats include adversarial prompt injection bypassing tool-calling guards or manipulating model routing to execute unauthorized actions.
Not certain from the listing — Arcade connects to files, emails, and calendars, but details on internal vector stores or RAG data operations are not specified. Threats include data exfiltration from connected sources via unauthorized tool queries.
Arcade provides a Custom Tool SDK and orchestrates tool-calling. Threats include insecure tool integration, tool misuse (e.g., unauthorized GitHub commits or Slack messages), and prompt injection leading to unintended API execution.
Supports 'Deploy Anywhere' (cloud, VPC, or on-premises). Threats include container compromise, credential theft from the hosting environment, and insecure storage of API keys/secrets.
Features 'Tool Evals' to automate and benchmark LLM-Tool interactions. Threats include blind spots in real-time execution monitoring and evaluation gaming where malicious tool calls bypass static benchmarks.
Features 'Agent Auth' to securely connect AI to services on behalf of users. Threats include token theft, session hijacking, privilege escalation, and authorization bypass within the authentication proxy.
Not certain from the listing — Arcade acts as a horizontal platform but does not explicitly detail a multi-agent marketplace. Threats include cascading failures if one connected service (e.g., Slack) triggers malicious actions in another (e.g., GitHub).
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).