Harpa AI — agentic threat model
Harpa AI presents a high-risk profile primarily due to its deployment as a browser extension with deep DOM access, session context awareness, and the ability to execute custom automation commands, making it a prime target for data exfiltration and session hijacking.
OWASP AIVSS score rationale
| Autonomy of Action | 0.60 | |
| Goal-Driven Planning | 0.40 | |
| Self-Modification | 0.10 | |
| Dynamic Tool Use | 0.70 | |
| Persistent Memory | 0.30 | |
| Contextual Awareness | 0.80 | |
| Dynamic Identity | 0.50 | |
| Multi-Agent Interactions | 0.10 | |
| 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.
Integrates with external foundation models (GPT, Claude). Threats include prompt injection via web page content (indirect prompt injection) leading to model reprogramming or mis-aligned outputs during summarization.
Processes active browser page data and performs automated data extraction. Threats include data exfiltration of sensitive session data, and knowledge-base poisoning if malicious web content is ingested during monitoring.
Orchestrates customizable automation commands and web monitoring. Threats include tool misuse, where malicious page elements trigger unintended automation commands or exploit insecure tool integrations.
Deployed as a browser extension. Threats include extension-level privilege escalation, unauthorized DOM access, and local storage compromise of API keys or session tokens.
Not certain from the listing — no details are provided regarding logging, guardrails, or drift detection for the browser-based automation and LLM interactions.
Not certain from the listing — no compliance certifications (e.g., SOC2), identity governance, or specific data privacy controls are detailed in the public directory.
Not certain from the listing — there is no mention of multi-agent coordination, agent-to-agent trust boundaries, or marketplace integrations.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).
These scores are auto-generated from public information (the agent's own listing, docs, and repository) using the canonical OWASP AIVSS formula and the MAESTRO framework — an estimate for guidance, not a penetration test, audit, or certification. See the scoring methodology. Are you the vendor? Factual corrections are free.