Tracked Capabilities for Safer Agents (martinodersky.substack.com)

2 points by odersky 3mo ago ↗ HN

1 comment

[ 2.9 ms ] story [ 17.8 ms ] thread

The blog post describes a new scheme to securing AI agents based on tracking capabilities in types. This got the best paper award at the inaugural ACM Conference on AI and Agentic Systems (CAIS 2026).