AWS open-sources Rust library to check AI agents’ tool calls
AWS has released Dogwood Local Engine (DLE), an open-source Rust library designed to control AI agents by checking their tool calls against user-defined temporal rules before they execute. The release addresses growing concerns about AI agents going out of control and taking unintended actions, providing what AWS describes as a safety "leash" for agent behaviour to prevent irreparable consequences from unchecked tool usage.
DLE works by receiving each tool call request, checking it against policies written in the Dogwood governance language (which AWS open-sourced in August), and returning a verdict to allow or deny the action. Testing showed DLE adds minimal delay to workflows – ranging from just 20 microseconds to 6 milliseconds depending on the policy window – while maintaining detailed event logs to preserve its state even if systems crash. An example use case involves restricting Git pushes to only occur after recent test runs pass within the preceding 15 minutes.
- AWS releases open-source tool to control AI agent actions with temporal rules.
- Performance impact minimal: adds 20 microseconds to six milliseconds delay.
- Addresses safety concerns about AI agents acting outside their intended limits.
New here? Start with this
Artificial intelligence agents are software systems that can independently perform tasks and make decisions with minimal human supervision. As organisations increasingly deploy these systems to automate complex work, they're giving them responsibility for critical operations.
AI agents can sometimes take unintended actions or behave in ways their users didn't anticipate, potentially causing irreversible harm. This is particularly concerning when agents can access sensitive systems or perform high-stakes operations like transferring money, deleting data, or modifying important records.
Given these risks, there has been growing focus on tools that can supervise and constrain AI agent actions before they happen. Such safety mechanisms work by examining each proposed action against predefined rules, allowing organisations to maintain control over automated systems.
AI Geopolitics Politics Technology
Read the full article at the source →
Originally published by The Register as “AWS offers local, open source leash for agent harnesses”.