Integrations
AI agent integration security
Integrations are where agent actions become real: an API call that refunds, sends, shares or deletes. These pages take one integration at a time, list the actions an agent can take through it using the vendor docs, show the abuse paths for each action, and describe what an Agent Action Pentest checks.
Surfaces we test. Names and logos belong to their owners and do not imply a customer relationship, partnership or endorsement.
Pages in this section
- Security testing for AI agents with Stripe accessRefunds, payouts and restricted API keys for agents that touch Stripe.
- Security testing for AI agents in SlackChannel history, message posting and link unfurling for agents installed in Slack.
- Security testing for AI agents with GitHub accessToken type, repository permissions and write paths for agents that touch GitHub.
- Security testing for AI agents with Gmail accessRead, send, modify and filter scopes for agents that work in a Gmail mailbox.
How these pages work
The model is rarely the thing that loses money or leaks data. The integration is. A support agent with a payments key, an email agent with send scope, a coding agent with a repository token: each one inherits whatever the credential allows. Our integration pages start from the vendor's own permission and scope documentation and work outward to what an attacker could make the agent do.
Each page lists the actions available through the API, with links to the endpoint or scope documentation. It gives the abuse path for each action and the VulnZap class it falls under. It quotes the least-privilege options the vendor documents, such as restricted keys or narrower scopes, and it includes an illustrative trace labeled as an example. Nothing on these pages describes a real customer finding.
Stripe came first, because refunds and payouts make the impact of an injected agent easy to see. Slack, GitHub and Gmail followed, each built from its own scope documentation. Salesforce, Zendesk, Google Drive and Jira are planned next.
Integration pages link back to the risks that drive them. Prompt injection supplies the instruction, excessive agency decides whether the credential can carry it out, and sensitive information disclosure covers what leaves when the integration can read customer or payment records.
Abuse paths are written as concrete sequences: what text reaches the agent, which API call it makes, and with which arguments. That format is deliberate. It is the same structure our reports use, so a page can be read as a preview of the kind of finding we would write up for that integration.
An integration test always runs against test mode, a sandbox or a staging account that you provide. We never need live payment keys or production mailboxes. If your agent connects to something not listed here, tell us at intake and we will agree scope before the engagement starts. The offer and checkout are below.
Agent Action Pentest: $2,500 fixed. Written report within 72 hours of access, or a full refund.
See the offer or email yaz@plaw.io
Questions
Do you need production credentials?
No. We test with test mode keys, sandbox accounts or staging credentials that you create for the engagement.
What if my agent uses several integrations?
We agree at intake which integrations are in scope and test the actions the agent can take through each of them.
Do you test the vendor API itself?
No. We test your agent and the code around its API calls, not the vendor platform.
Related pages
- AI agent security risksAI agent security risks with OWASP and MITRE ATLAS references: prompt injection, excessive agency, data disclosure and MCP tool poisoning, and how we test.
- AI agent security standards, mappedOWASP Top 10 for LLM Applications, OWASP Agentic AI threats and MITRE ATLAS, each mapped to what an Agent Action Pentest tests and what it leaves out of scope.
- AI agent framework securityHow AI agent frameworks like the OpenAI Agents SDK and LangChain expose tools, handoffs and guardrails, which risks apply, and what an agent pentest checks.
- AI agent security testing, comparedHow an agent action pentest compares with other AI security testing, such as AI red teaming, using public definitions only, and when each is the better fit.
Sources
Last reviewed: