# [Live Event] CoreBreak, explained: 2 ways past AI agent guardrails

Canonical: https://brew.new/templates/redpanda/live-event-corebreak-explained-2-ways-past-ai-agent-guardrails

Brand: redpanda.com
Category: general

![Preview of [Live Event] CoreBreak, explained: 2 ways past AI agent guardrails](https://cdn.brew.new/email-preview-62b7ff5693cf8c57-tracking_r57j8ba6rxxwt3n602wf6sn1958gdf2q-1792167690502.png)

## Email content

CoreBreak: What a Real AI Agent Flaw Proves About Where Guardrails Belong

Hi there,

There were two distinct ways in. One involved faking a tool call. The other involved faking a human approval that no person ever actually saw.

Both worked because every guardrail in play — prompts, content filters, refusal training — was built around the model itself. Skip the model, and none of it fires.

This Tuesday, Tyler Akidau and Marc Millstone walk through exactly how both attacks worked, and why the same flaw showed up independently across three of the industry’s biggest agent platforms.

Tue, 10/20 at 9:30AM PT.

Register Now

See you there,

Team Redpanda

twitter-1

slack-1

github-1

linkedin-1

Youtube

Youtube

© 2026 Redpanda Data, Inc.

220 Montgomery St. #400, San Francisco, CA 94104

Manage Preferences

[Open and remix this design](https://brew.new/templates/redpanda/live-event-corebreak-explained-2-ways-past-ai-agent-guardrails)

[Browse email designs](https://brew.new/browse/templates)