Threat
If Agents Were Angels, No Governance Would Be Necessary:Out-of-Band Policy Enforcement at a Trusted Tool Boundary
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 24 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
CoreBreak’s reported bypass highlights the need to enforce tool authorization outside agent reasoning rather than relying solely on model guardrails to reject attacker-controlled calls.
Ground News reports that CoreBreak lets attacker-authored tool calls bypass in-model guardrails in AWS Bedrock AgentCore (Strands SDK), Google’s Agent Development Kit, and Vercel’s AI SDK, with vendor patches reportedly issued. The linked Redpanda paper separately evaluates out-of-band tool-policy enforcement, reporting trace failures falling from 57.6% to 0.2% in controlled trials. The provided paper excerpt does not independently substantiate the platform-specific CoreBreak claims.