Korven
AI agents can act. But they have zero security
2 followers
AI agents can act. But they have zero security
2 followers
We require HTTPS for web requests, OAuth for APIs, 2FA for logins. But AI agents? Zero identity. Zero authorization. Zero kill switch. AIP is the missing security layer: Ed25519 identity per agent (DID-based) Cryptographic boundary enforcement - not prompt guardrails Instant global revocation - one API call <1ms verification, deterministic (not probabilistic) @observe - free agent debugging (NEW v0.4.0) Open-source. Python + TypeScript. Works with LangChain, CrewAI, AutoGen.
📌 Hey Product Hunt! 👋
During a demo, I watched an AI agent nearly drain a test account. In 3 seconds it
had authorized a $50K transfer with zero identity checks.
That's when I realized: we have HTTPS for web traffic, OAuth for APIs, 2FA for logins —
but AI agents execute actions with nothing but a system prompt and hope.
So I built AIP - the cryptographic protocol layer that was missing:
@shield(passport, actions=["transfer_funds"], limit=500)
One decorator. Every call is Ed25519-signed, boundary-checked, and revocable instantly.
What shipped today (v0.4.0):
• @observe - free agent visibility (logs everything, blocks nothing)
• One-line upgrade: @observe → @shield when you need enforcement
• 98 tests passing, Python + TypeScript SDKs
This is NOT prompt guardrails. It's deterministic, cryptographic enforcement,
the difference between hoping your agent follows instructions and knowing it can't
violate them.
I'd love to hear: what would make YOU trust an AI agent in production?
pip install aip-protocol
npm install aip-protocol
https://korven.cc
GitHub: https://github.com/theaniketgiri/aip