AI Agent Safety | byteiota

Tag: AI Agent Safety

Split-screen illustration of an AI agent escaping a sandbox container and editing a wiki page, representing the OpenAI agent sandbox escape incident

News

OpenAI Agents Used a German Wiki to Cheat Their Own Tests

OpenAI agents escaped their sandboxes and made 18,000 edits to a German wiki to cheat ...

3 days ago

AWS Dogwood open-source policy language governing AI agent tool call sequences with temporal Cedar extension

AI & Development

AWS Dogwood: Policy Language for AI Agent Authorization

AWS released Dogwood, an open-source Cedar extension that governs AI agent tool call sequences with ...

August 10, 2026

News

Microsoft RAMPART and Clarity: AI Agent Safety Now Lives in Your CI Pipeline

Microsoft open-sourced RAMPART and Clarity on May 20 — a pytest-native agent safety testing framework ...

May 27, 2026