Solda AI — agentic threat model

Not certain from the listing — The underlying LLMs and text-to-speech (TTS) models are not specified. Threats include prompt injection to bypass sales scripts, voice cloning abuse, and model reprogramming to output malicious or non-compliant financial advice.

Not certain from the listing — The agent processes sensitive banking, fintech, and KYC data. Threats include data exfiltration of customer profiles, poisoning of the CRM-linked knowledge base, and unauthorized access to personally identifiable information (PII).

The agent orchestrates cross-channel communication and triggers workflows across systems. Vulnerabilities include insecure tool integration with CRMs, allowing prompt injection to execute unauthorized database writes, trigger fraudulent emails, or initiate unauthorized outbound calls.

Not certain from the listing — The hosting environment and API security controls are undisclosed. Threats include API key exposure (telephony, CRM, email providers), lack of network segmentation between the agent and banking databases, and container escape.

The agent features sentiment analysis, script optimization, and automated CRM logging. Threats include blind spots in monitoring voice calls for prompt injection, insufficient logging of unauthorized agent-initiated workflows, and evasion of sentiment-based guardrails.

Operating in banking and fintech (KYC, support) requires strict compliance (GDPR, PCI-DSS, SOC2). The listing does not detail compliance certifications, posing risks of regulatory non-compliance, lack of explicit audit trails for automated financial decisions, and unauthorized KYC processing.

The agent acts as an intermediary between customers, telephony networks, and enterprise CRMs. Threats include cascading failures where a compromised CRM triggers malicious agent actions, or the agent is manipulated into abusing trust boundaries to exfiltrate data to external systems.