“Existing endpoint and API security tools are ineffective at securing AI agents because they lack the context to understand the agent's intent or reasoning.”