Overview
The Fireblocks Genie is an assistive interface that helps customers explore treasury data, generate insights, and draft recommended actions.
- No autonomy: The agent does not execute transactions, change configurations, or bypass user permissions. It cannot perform write operations without an explicit user‑approved action via the standard Fireblocks UI or an approved automation flow.
- No new permissions: The agent inherits the customer’s existing roles and permissions. If a user cannot access a dataset or action today, the agent cannot either.
- Not advice: Outputs are for informational purposes only and are not financial, legal, tax, accounting, or compliance advice. Always verify before acting.
- Privacy: Fireblocks does not use customer content to train third‑party models. Provider access is limited to contracted, allow‑listed services and governed by Fireblocks’ privacy and security practices.
- Limitations: Large language models can be incorrect or incomplete. Use human judgment, and report issues to Support.
General AI Policy & Scope
What is Fireblocks’ stance on AI usage?
Fireblocks responsibly integrates AI to improve product usability and operational efficiency, with security, privacy, and compliance as primary design constraints.
What does the Fireblocks AI Agent do?
It enables natural‑language querying of treasury data, contextual explanations, and drafting of recommended next steps. It is a read‑first, assistive experience and does not execute changes without user initiation.
Which technologies power the agent?
The agent uses Anthropic Sonnet (enterprise) + MCP (Model Context Protocol) to orchestrate tools. MCP selects and invokes allow‑listed tools such as Snowflake Cortex (for secure compute over curated data) and Fireblocks APIs (subject to the user’s RBAC and environment policies).
Use of Snowflake Cortex & Anthropic Sonnet
What is Snowflake Cortex used for?
Snowflake Cortex securely processes and hosts conversational AI in Fireblocks. It ensures that customer queries are handled within a trusted, enterprise-grade environment.
What is Anthropic Sonnet used for?
Anthropic Sonnet interprets natural language questions and translates them into Fireblocks API calls, powering conversational capabilities such as the MCP (Model Context Protocol).
Are there any restrictions or guidelines for using these models?
Yes. Usage must comply with the AI Services terms of the Fireblocks’ terms and conditions and the acceptable use policies of the providers mentioned which are available on their websites: Anthropic and Snowflake.
Both models are operationally restricted to:
- Treasury-related queries
- Fireblocks APIs and data sources
- Providing insights or guidance within the Fireblocks console
They are not used for personal profiling, marketing, or external data scraping.
Security & Data Privacy
How do these AI models interact with sensitive customer data?
Customer data remains protected under Fireblocks’ security and privacy framework. Data submitted by the customer and/or its users will be processed in accordance with the DPA available here and the Fireblocks Privacy Policy, available here.
The AI agent has no ability to store or learn from data across sessions, and no data is shared with third parties outside contracted, compliant providers.
How do we prevent AI models from "learning" or "leaking" sensitive information?
- No fine-tuning with customer data
- Strict data access controls
- Session-level isolation
- Regular monitoring for data leakage risks
What measures are in place to secure AI outputs?
AI responses are read-only recommendations. Human users must review and approve any actions, ensuring full control remains with the customer.
Is there human oversight of AI usage?
Yes. AI outputs are subject to Fireblocks’ review mechanisms, and customers always maintain control of all final decisions and actions.
Responsible AI & Future Development
What is Fireblocks' philosophy on responsible AI?
We prioritize accuracy, transparency, fairness, and accountability. All AI deployments undergo internal reviews for security, compliance, and business alignment.
Can I disable the Fireblocks Genie app feature?
Yes, in order to disable the Fireblocks Genie admins can directly disable Genie from the settings.
How can I provide feedback or raise concerns?
Customers can share feedback directly in the Fireblocks Genie app using the thumbs up/down or Report button with free text. Any concerns about responses or security are promptly reviewed and escalated to the appropriate teams.