AI Term Highlighter — agentic threat model

Not certain from the listing — the specific LLM backend is undisclosed. However, the model is highly vulnerable to indirect prompt injection, as it processes arbitrary webpage text which could contain malicious instructions designed to hijack the definition generation or manipulate the extension's behavior.

Not certain from the listing — the data pipeline, transmission protocols, and caching mechanisms for webpage content are unspecified. The primary threat is data exfiltration, where sensitive user data or PII present on scanned webpages is inadvertently transmitted to the LLM provider or backend servers.

Not certain from the listing — the orchestration framework is not detailed. The main threat is insecure tool integration, specifically how the extension interacts with the browser DOM to highlight terms and inject definitions, which could be exploited to execute arbitrary scripts (DOM-based XSS).

Not certain from the listing — the hosting environment for the definition API is unknown. As a Chrome extension, it operates within the browser sandbox, but overly permissive manifest permissions (e.g., '<all_urls>') could allow a compromised extension to access sensitive data across all open tabs.

Not certain from the listing — there is no mention of content filtering, input sanitization, or output guardrails. This creates a blind spot where toxic, inaccurate, or malicious definitions generated by the LLM are rendered directly to the user without validation.

Not certain from the listing — compliance with data privacy regulations (like GDPR/CCPA) is unclear, especially regarding the processing of user-browsed content. The closed-source nature of the extension prevents independent security audits of its data handling practices.

Not certain from the listing — the extension does not appear to interact with other agents or marketplaces. The primary ecosystem risk is limited to supply chain attacks on the Chrome Web Store listing or dependencies used within the extension.