Sweet Security: AI Agent Blocking for Cloud Security

An Israeli startup called Sweet Security has introduced a new feature to stop artificial intelligence agents from performing unauthorized actions. This capability, called Agentic AI Blocking, halts AI agents mid-execution if they stray from their intended functions. What's interesting is this extends Sweet Security's current enforcement tools for cloud workloads to the AI agents working alongside them. The goal is to prevent unauthorized actions before they can cause damage. This new system aims to block unauthorized tool calls and agent sessions as they happen, stopping sensitive data like secrets and personal information from being stolen. It also works against prompt injection attacks that could redirect an agent's behavior. The system uses a runtime reasoning layer that analyzes over one billion runtime events daily to learn normal behavior. Actions outside these norms are flagged as unauthorized. This helps enterprises adopt AI technologies with more confidence by immediately blocking malicious or unintended agent activities.