NameTrap: Steering AI Coding Agents Toward Security Vulnerabilities via Variable Names

Evidence record

Identifier-only manipulation tests across six coding agents and CWE-specific code-generation tasks. The malicious semantic change appears in the agent completion, not in preexisting exploit code.

Read the original source ↗

Claim-level citations (2)

• supportsNameTrap induced vulnerable code from identifier-only source changes: Across six coding agents under the study’s matched setup, overall attack success ranged from 22.5% to 56.0% for identifier-based bait in selected CWE categories.

  Section 4, Table 2; six-agent comparison

• supportsNameTrap induced vulnerable code from identifier-only source changes: The test covers selected command injection, XSS, cryptographic and deserialization weakness classes; malicious semantics were introduced in later agent output.

  Section 3 threat model; identifier-only verification