Toxic Combinations: When Cross-App Permissions Stack into Risk

When AI agents bridge multiple applications through OAuth or MCP connections, they create hidden permission combinations that single-app security reviews cannot detect.

• •Moltbook's January 2026 breach exposed 1.5 million AI agent API tokens and 35,000 email addresses stored with unencrypted plaintext credentials like OpenAI API keys.
• •Toxic combinations form when integrations connect two or more apps through tokens and scopes no single app owner authorized as a combined risk surface.
• •Conventional access reviews examine apps independently and miss runtime trust relationships created by agents, integrations, and bot connectors.
• •Organizations need cross-app scope reviews, non-human identity inventories, and runtime drift monitoring to catch toxic combinations.
• •