Tidio — agentic threat model

Not certain from the listing — the specific foundation models powering Lyro are not disclosed. However, the primary threat at this layer is adversarial prompt injection, which could allow users to bypass customer service guardrails, extract system prompts, or generate inappropriate content.

Not certain from the listing — the exact database and vector store architecture for RAG are unspecified. The agent accesses Shopify order history and cart data, making it vulnerable to data exfiltration or unauthorized PII retrieval if the data access controls are not strictly isolated per session.

Not certain from the listing — the underlying orchestration framework is proprietary. The integration with Shopify APIs for cart previews and order history lookups introduces risks of tool misuse, where a manipulated agent could execute unauthorized API calls or leak order details.

Not certain from the listing — details regarding cloud hosting, network isolation, and API credential storage are omitted. Compromise at this layer could expose Shopify API keys or social media integration tokens, leading to broader platform compromise.

Not certain from the listing — the presence of real-time guardrails or automated evaluation systems is not detailed. Without robust observability, malicious prompt injections or hallucinated product recommendations could go undetected until reported by users.

Not certain from the listing — compliance certifications (such as SOC 2 or GDPR compliance details) are not explicitly mentioned. Strong identity and access management are critical to ensure that customer support agents and the AI bot itself only access authorized tenant data.

Not certain from the listing — while the agent integrates with Shopify and social media platforms, it is unclear if it interacts with other autonomous agents. The primary ecosystem risk is cascading trust, where a compromise in Tidio could be leveraged to exploit connected Shopify stores.