Skip to content
AgentProv

2026-05-21 · 2026-Q2

Composio Agent Integration Platform Breach Exposes 10,000+ Customer Credentials

Primary layerL1 CredentialSecondary:L3 ToolL5 Config

Should this run hold this key at all?

The question layer 1 asks, and failed here.
Fixable
Unknown
Vendor response
Unknown
In the wild
Yes
Actor
Attacker

Target

Composio, an integration platform that brokers third-party tool access for AI agents

Impact

Attackers exfiltrated 5,241 API keys and 5,001 GitHub OAuth tokens, plus 12 Gmail tokens and smaller numbers of Jira, Slack, HubSpot, Linear, Notion, and Google Calendar tokens, for a total above 10,000 customer credentials. Because Composio's role is to hold downstream credentials on behalf of customer agents, a single breach of the broker handed the attacker working access to the connected services of every affected customer

Root cause

A compromised Gmail OAuth token belonging to a Composio employee gave the attacker inbox access, which was used to intercept magic-link sign-in emails. From there the attacker registered malicious tool definitions to obtain arbitrary code execution in the sandbox and reached the production credential cache. The same OAuth-token pivot pattern as the Vercel and Context.ai chain a month earlier

Sources

Catalog

Cite this: AgentProv incident record: Composio Agent Integration Platform Breach Exposes 10,000+ Customer Credentials (2026-05-21). https://agentprov.com/incidents/2026-05-21-composio-agent-integration-platform-breach-exposes-10-000-customer-credentials