Explore 2 source-supported AI security cases for 13.28 · UI/UX Deception in Databricks DASF 3.0, with evidence, scope, and defensive analysis.
Risk category
The MCP client interface misleads users about server capabilities, permissions, or actions. Hidden tool calls, unclear descriptions, and misleading permission dialogs can keep users from understanding what they are approving. A user may consequently authorize a dangerous operation while believing it is a different or less consequential action.
pgEdge disclosed independent SQL-injection, read-only-enforcement and bundled-client approval defects affecting versions through 1.0.0 and repaired in 1.1.0. A custom pl-do tool could turn argument data into SQL; query_database could defeat an application-level read-only setting; two client defects skipped write confirmation. Database permissions still bound the resulting actions. The source establishes controlled validation, not a malicious customer breach.
Why it relates: The bundled clients omit the expected confirmation for disguised write queries and write-capable custom tools. This is an absent approval interaction, not a demonstrated forged dialog or successful human deception.
Each advisory contributes its own route and prerequisites; client confirmation and database authorization are different controls.
Check Point researchers tested how Claude Code, an AI coding assistant, loads files from an untrusted repository. Separate hook, MCP startup and API-base-URL paths produced local command execution or API-key disclosure. The hook path required accepting repository trust; the other two could act before that decision. The demonstrated workspace-file access was limited to the tested key and workspace.
Why it relates: The MCP approval path fails to provide the intended independent decision before a project-supplied server command is launched.
Three native project-startup paths, with the tested credential consequence. Model prompt interpretation is not the exploit.