Periscope Chat — agentic threat model

Not certain from the listing — likely utilizes commercial LLMs (e.g., OpenAI, Anthropic) to handle multilingual conversations. Primary threats include prompt injection that could hijack the brand voice or bypass safety guardrails.

Not certain from the listing — likely uses RAG to store and retrieve machine operation guidance and maintenance steps. Threats include knowledge-base poisoning, which could result in the agent outputting dangerous or incorrect physical maintenance instructions.

Orchestrates conversational flows and triggers integrations with Google Sheets, Calendars, and ERPs. Threats include insecure tool execution where malicious user inputs trigger unauthorized API calls or data exfiltration via connected integrations.

Not certain from the listing — deployed as a closed-source SaaS platform. Threats include insecure storage of third-party API credentials (ERPs, Google, Slack) and lack of network isolation between tenant environments.

Provides a Conversation Dashboard for human oversight, sentiment tracking, and live summaries. Threats include blind spots in automated sentiment analysis and the potential for attackers to bypass detection mechanisms during malicious interactions.

The 'no logins required' feature for factory floor workers introduces significant authentication and authorization risks, potentially allowing unauthorized users to access sensitive operational data or trigger ERP actions without accountability.

Not certain from the listing — operates primarily as a single-agent platform connecting to external APIs rather than a multi-agent ecosystem. Threats are limited to cascading failures if connected third-party APIs or ERP systems are compromised.