Your AI agent shouldn't surprise you.
No "it decided to delete the database."
No "I didn't tell it to do that."
Codios = cryptographic contracts for every agent-to-agent call.
What you authorize is exactly what happens. Nothing more.
No surprises.
https://t.co/9M9FHM8FEZ
@lifeof_jer The agent knew the rule. It violated it anyway. That's not an AI problem — that's an authorization problem.
Signed capability contracts make volumeDelete literally uncallable without explicit human-granted scope.
We built Codios for that. Try it → https://t.co/zDwuHDpzIx
Codios - AI Agent Caging Infrastructure: Signed capability contracts for AI agents
Officially on Product Hunt today.
https://t.co/8SpbrpIhS7 via @producthunt
We give every agent a crypto identity — Ed25519 keypair + DID:key.
Before Agent A can call Agent B, it shows a signed contract:
- caller DID
- callee DID
- allowed scopes (eg, payment:charge:max_10Kusd)
- expiry + nonce (replay protection)
Verified locally. Zero network latency
Most multi-agent systems work like this:
Order Agent → calls → Payment Agent
The payment agent processes it because it arrived on port 8080. That's it.
No identity check, no scope validation, no replay protection.
Fine in dev. A liability in prod.
@manuelsampedrop That's the right start. The missing piece: proving authorization at the moment of action. A log doesn't prove scope or replay.
We add signed contracts to every call, Ed25519 over action+scope+nonce. Audit log becomes cryptographically verifiable. Codios handles this now.
How do you audit what your AI agent actually did?
Does anyone have a clean way to enforce and audit agent actions without adding a ton of overhead?
Curious how others are handling non-repudiation for autonomous agents.
@DavidLinthicum Yeah, agent-data interaction boundaries are key. I personally use Codios to define what my agents can access and to audit their actions, makes it way easier to sleep at night.
@cybernewslive Great resource. For agentic AI, we have tools like Codios to cryptographically restrict agent permissions and keep audit trails. Makes it easier to enforce the kind of controls OWASP suggests.
@couchbase@InfoWorld Great list! For the permission enforcement part, we have a tool that lets you cryptographically sign agent capabilities, makes auditing way easier.
We give every agent a cryptographic identity — Ed25519 keypair + DID:key.
Before Agent A can call Agent B, it presents a signed contract:
- caller DID
- callee DID
- allowed scopes (e.g., payment:charge:max_10000usd)
- expiry + nonce (replay protection)
Zero network latency.