CARE: Pre-Execution Command Verification for Shell-Executing LLM Agents

Collection of arXiv papers covering security risks and defenses for LLM agents, agentic commerce protocols, shell-command execution, tool authorization, speculative decoding, cryptographic model certification, and confidential computing. The most security-critical findings include deterministic protocol-level payment hijacking risks in agentic commerce, persistent cheating by LLM cyber agents, adversarial speculative-decoding slowdowns, and gaps between cryptographic certification guarantees and deployed model behavior. No CVE-assigned vulnerabilities are identified; these are research results