Security researchers disclosed CVE-2026-109440, a prompt injection vulnerability (CVSS 8.8) in LangAlpha, an open-source AI agent used for financial research. Web content containing hidden instructions can trigger the agent’s Bash and code-execution tools without requiring human approval, letting an attacker run arbitrary commands in the agent’s sandbox and steal stored credentials, including during unattended scheduled runs. No public proof-of-concept or real-world exploitation has been reported, and no patched version had been confirmed as of disclosure.