TokenVetoUse cases
Wherever AI acts, permissions must hold
TokenVeto tests the four architectures where permission failures hurt most — with evidence, not vibes.
Customer-facing chatbots
Support bots connected to order systems, CRMs and refund tools can be manipulated into actions far beyond answering questions.
Real outcome
A fintech support bot passed 412 of 438 tests; the 26 failures included a refund path reachable without authentication. Fixed before launch.
What we test
- ▸Prompt injection via user messages and uploaded files
- ▸Unauthorized refunds, discounts and account changes
- ▸Cross-customer data leakage between sessions
- ▸Tool-call abuse through chained instructions
Internal copilots & assistants
Copilots with access to email, docs and calendars inherit every employee's permissions — and every employee becomes a potential attack surface.
Real outcome
An enterprise copilot rollout was paused after TokenVeto showed a crafted email could make the assistant forward confidential contracts.
What we test
- ▸Indirect injection through emails and shared documents
- ▸Privilege escalation across departments
- ▸Exfiltration of HR, finance and legal data
- ▸Action approval bypass (send, delete, share)
Autonomous agents
Agents that plan and act over many steps accumulate small permission slips into major incidents — often without any human in the loop.
Real outcome
A coding-agent vendor adopted weekly TokenVeto runs after tests showed the agent could be steered to exfiltrate repository secrets.
What we test
- ▸Multi-step goal hijacking
- ▸Sandbox and tool-scope escape attempts
- ▸Memory poisoning across sessions
- ▸Unsafe chaining of approved actions
RAG & knowledge pipelines
Retrieval systems trust their corpus. Poisoned or malicious documents turn the knowledge base itself into an attack vector.
Real outcome
A legal-tech platform discovered confidential client documents were retrievable across tenant boundaries in 3 of 12 test scenarios.
What we test
- ▸Document-level prompt injection
- ▸Tenant isolation between data sources
- ▸Access-control enforcement on retrieved chunks
- ▸Citation spoofing and source manipulation
Built for your industry
Test scenarios tuned to the data and regulations that matter to you.
Finance & Insurance
Refund abuse, transaction authorization, KYC data exposure
Healthcare
Patient record isolation, diagnosis-assistant boundaries
SaaS & Developer Tools
Agent tool-scope, repository and secret access
Retail & E-commerce
Discount manipulation, order data leakage
Legal & Compliance
Tenant isolation, privileged document access
Public Sector
Citizen data boundaries, audit-ready evidence