Tanka — agentic threat model
Tanka presents a high data-security risk profile due to its core feature of long-term enterprise memory and tool integration, which increases the impact of memory poisoning and data exfiltration if compromised.
OWASP AIVSS score rationale
| Autonomy of Action | 0.40 | |
| Goal-Driven Planning | 0.30 | |
| Self-Modification | 0.20 | |
| Dynamic Tool Use | 0.50 | |
| Persistent Memory | 0.90 | |
| Contextual Awareness | 0.80 | |
| Dynamic Identity | 0.20 | |
| Multi-Agent Interactions | 0.30 | |
| Non-Determinism | 0.60 | |
| 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 — The underlying foundation models are not specified, leaving the system vulnerable to standard LLM threats like adversarial prompt injection and model-specific alignment bypasses.
Tanka heavily relies on long-term memory and learning from past conversations, making it highly susceptible to data/knowledge-base poisoning and unauthorized data exfiltration of sensitive enterprise context.
The agent framework manages long-term memory and tool integrations. Insecure tool integration and memory poisoning are primary threats, as malicious chat inputs could manipulate the agent's stored context and tool execution.
Not certain from the listing — No deployment details, hosting environments, or sandboxing mechanisms are described, presenting potential risks of container compromise or unauthorized lateral movement.
Not certain from the listing — There is no mention of evaluation frameworks, real-time monitoring, or guardrails to detect drift, anomalous memory updates, or malicious inputs.
Not certain from the listing — Compliance certifications (e.g., SOC2, GDPR) and enterprise access controls are not detailed, which is a critical gap for a tool acting as an 'enterprise brain'.
Not certain from the listing — While designed for team chat environments, it is unclear if Tanka interacts directly with other autonomous agents or if it operates solely as a single-agent assistant.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).