OpenAdapt — agentic threat model
OpenAdapt presents a high-risk profile due to its ability to record and replay arbitrary desktop GUI actions, essentially acting as a local automation engine with access to sensitive user inputs. While its built-in PII/PHI scrubbing mitigates some data exposure risks, the potential for visual prompt injection or malicious replay actions on the host system remains a critical concern.
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 | 0.60 | |
| Contextual Awareness | 0.80 | |
| Dynamic Identity | 0.30 | |
| Multi-Agent Interactions | 0.10 | |
| 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.
Uses Visual Language Models (VLMs) and Segment Anything for GUI understanding. Vulnerable to visual prompt injection (e.g., malicious instructions embedded in web pages or desktop applications being automated) and adversarial GUI elements that could hijack the model's execution flow.
Captures screenshots, mouse/keyboard inputs, and tokenized formats. While it includes PII/PHI scrubbing, any failure in this pipeline could lead to the exposure of highly sensitive credentials, personal data, or proprietary information in recording logs and process graphs.
Orchestrates recording, process graph generation, and synthetic input replay. Insecure tool integration is a major risk here, as the agent translates model outputs directly into OS-level mouse and keyboard events, potentially executing destructive actions if the planning logic is manipulated.
Runs locally on desktop environments (virtualized, web, and native). If the agent is compromised, it acts as a local remote-access tool, allowing arbitrary interaction with the host OS without sandboxing boundaries mentioned in the listing.
Provides visualization of user inputs, recordings, and process graphs. However, there is a lack of real-time guardrails or anomaly detection to intercept malicious or unintended synthetic inputs before they are executed on the desktop.
Features privacy-focused PII/PHI scrubbing as a core compliance control. However, because it is open-source and runs locally, centralized access controls, audit logging, and enterprise policy enforcement are left to the user's infrastructure.
Not certain from the listing — The description focuses on standalone desktop and web process automation. There is no explicit mention of multi-agent orchestration, marketplace integrations, or agent-to-agent communication protocols.
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.