Mirtilla — agentic threat model
Mirtilla is a low-autonomy productivity agent focused on meeting transcription and note-taking. Its primary security risks center around data privacy and indirect prompt injection via meeting audio/transcripts, mitigated partially by its advertised encrypted storage.
OWASP AIVSS score rationale
| Autonomy of Action | 0.20 | |
| Goal-Driven Planning | 0.10 | |
| Self-Modification | 0.00 | |
| Dynamic Tool Use | 0.20 | |
| Persistent Memory | 0.40 | |
| Contextual Awareness | 0.50 | |
| Dynamic Identity | 0.00 | |
| Multi-Agent Interactions | 0.00 | |
| Non-Determinism | 0.50 | |
| Opacity & Reflexivity | 0.60 |
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 — likely relies on external foundation models (e.g., Whisper for transcription, GPT/Claude for summarization). Main threats include indirect prompt injection embedded in meeting audio/text and data leakage to upstream model providers.
The agent stores meeting transcripts and notes using 'fully encrypted' storage. Key threats include cryptographic key mismanagement, unauthorized access to transient unencrypted data during processing, and data exfiltration.
Not certain from the listing — orchestration is likely a simple pipeline from audio-to-text to summarization. Threats include insecure handling of 'Custom AI Requests' which could allow users to bypass system prompts or manipulate outputs.
Not certain from the listing — likely hosted on standard cloud infrastructure. Threats include insecure API endpoints, lack of isolation during audio file processing, and server-side request forgery (SSRF) if custom requests fetch external data.
Not certain from the listing — no observability or guardrail mechanisms are mentioned. Threats include a lack of monitoring for malicious injections in meeting transcripts or drift in summarization quality.
The service implements data encryption at rest ('fully encrypted'). However, there is no mention of access control mechanisms, multi-tenant isolation, or compliance certifications (e.g., SOC2, GDPR) which are critical for handling sensitive meeting data.
Not certain from the listing — the agent appears to operate as a standalone service with no multi-agent or ecosystem integrations described.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).