Etodist — agentic threat model

Not certain from the listing — likely utilizes multimodal foundation models for video, audio, and text processing. Primary threats include indirect prompt injection via classroom dialogue (e.g., students or teachers speaking malicious instructions to manipulate the evaluation) and inherent model biases in pedagogical assessment.

Not certain from the listing — ingests and processes classroom video and audio recordings. Key threats include unauthorized access to or exfiltration of sensitive student and teacher data, data poisoning of the comparative analysis database, and lack of secure retention/deletion policies for media files.

Not certain from the listing — orchestration likely involves sequential pipelines for transcription, metric extraction, and report generation. Threats include insecure processing of large media files and potential manipulation of the 25-metric evaluation logic via prompt injection.

Not certain from the listing — as an open-source and paid platform, deployment could be self-hosted or cloud-managed. Threats include insecure cloud storage buckets containing raw classroom recordings and standard web application vulnerabilities (e.g., broken object-level authorization).

Not certain from the listing — provides objective feedback but does not detail how it monitors for evaluation gaming (e.g., teachers adapting their speech specifically to trick the 25 metrics) or how it detects hallucinations in the generated recommendations.

Not certain from the listing — processing classroom recordings of minors strongly implicates strict regulatory frameworks like FERPA, COPPA, and GDPR. The listing does not specify compliance certifications, access controls, or consent management workflows.

Not certain from the listing — appears to operate as a standalone analytical tool with no multi-agent coordination or external marketplace integrations mentioned, resulting in minimal ecosystem-level threats.