Askiva — agentic threat model

Not certain from the listing — the underlying foundation models are not specified. However, conducting live, unstructured interviews in 10+ languages exposes the model to adversarial prompt injection by participants, potentially leading to mis-aligned outputs, brand damage, or system reprogramming during the session.

Not certain from the listing — the storage mechanism for participant lists, transcripts, and recordings is not detailed. The primary threat is data exfiltration of sensitive participant PII and proprietary research insights, as well as potential knowledge-base poisoning if summaries are fed back into a shared repository.

Not certain from the listing — the orchestration framework is proprietary. Key threats include insecure tool integration with scheduling and email systems, and memory poisoning if a malicious participant's input influences the agent's behavior in subsequent interviews.

Not certain from the listing — hosting, sandboxing, and infrastructure security details are omitted. Threats include unauthorized access to the hosting environment, which could compromise API keys used for scheduling, email dispatch, and transcription services.

Not certain from the listing — there is no mention of real-time guardrails, logging, or observability tools to monitor live interviews. This creates a blind spot where abusive, manipulative, or off-topic behavior by the agent or participant may go undetected until after the interview is complete.

Not certain from the listing — compliance certifications (such as GDPR, which is highly relevant for recording and storing participant interviews) and access controls are not mentioned, representing a significant regulatory and privacy risk.

Not certain from the listing — there is no indication of multi-agent coordination or integration with an external agent ecosystem, meaning cascading agent-to-agent failures are currently a low risk.