Most AI safety tools use prompt-level filters:

- "Don't do anything harmful"

- LLM-as-judge (another model watching the first one)

- Retrieval-based guardrails

We've all seen it - an agent goes rogue during a demo, a test agent

accesses production data, a chatbot sends emails it shouldn't.

What's the worst thing an AI agent has done in your system?

(Bonus points if it happened in production )

We just shipped v0.4.0 with @observe, a free, lightweight way to

see what your agents are doing.

from aip_protocol import passport, observe

Honest question for the community:

We require HTTPS for every web request. We require OAuth for every API call. We require 2FA for every login.

But AI agents? We let them execute actions with nothing more than a system prompt and a prayer.

An agent can transfer money, delete data, email customers and the only "security" is hoping the LLM follows instructions. Prompt injection can bypass that in one message.