Evidence · DiggingBeagle record
CARI: Cross-Boundary Adversarial Runtime Injection in Coding-Agent Workflows
CARI reports 2,845 successful attack trials out of 3,000 in a controlled mini-swe-agent setup that injected malicious runtime feedback, with separate product-agent configurations yielding lower success.
- Source role
- primary disclosure
Evidence record
CARI reports 2,845 successful attack trials out of 3,000 in a controlled mini-swe-agent setup that injected malicious runtime feedback, with separate product-agent configurations yielding lower success.
Claim-level citations (1)
- supportsCARI turns runtime diagnostics into unauthorized coding-agent actions: CARI reports 2,845 successful attack trials out of 3,000 in a controlled mini-swe-agent setup that injected malicious runtime feedback, with separate product-agent configurations yielding lower success.
Threat Model, Runtime-Feedback Construction and Evaluation