CyBot — agentic threat model

Not certain from the listing — likely utilizes proprietary or third-party speech-to-text and NLU/LLM models. Threats include adversarial voice inputs (audio prompt injection), model reprogramming, and misaligned or toxic voice outputs during live customer calls.

Not certain from the listing — processes real-time voice streams and transcripts. Threats include the exfiltration of sensitive customer PII spoken during calls, lack of secure transcription storage, and potential data poisoning if call logs are used for downstream fine-tuning.

Not certain from the listing — orchestrates conversation flows and context-awareness. Threats include voice-based jailbreaking to bypass business logic, and insecure tool integration if the agent connects to CRMs or booking systems to execute actions.

Not certain from the listing — hosted on Cyfuture's enterprise-grade infrastructure. Threats include SIP/RTP stream interception, unauthorized API access to the voice gateway, and lack of sandboxing for session handling.

Not certain from the listing — claims high accuracy and performance but does not detail real-time guardrails. Threats include undetected hallucinations where the bot makes unauthorized financial or contractual commitments to customers.

Not certain from the listing — mentions 'robust data security' but lacks explicit compliance certifications (e.g., PCI-DSS for handling credit card details over phone, SOC2, or HIPAA).

Not certain from the listing — operates primarily as a standalone horizontal voice agent. Threats of multi-agent cascading failures are low unless integrated into a larger multi-agent customer service ecosystem.