Cryptographic trust
for autonomous
AI agents.
A minimal protocol to make agent actions verifiable and tamper-evident.
No blockchain. No consensus. No central authority.
agent = AgentIdentity.create()
event = agent.stamp("user_prompted", {"data": "hello"})
assert agent.verify(event)
# → signed, hash-chained, tamper-evident
Logs can be modified
No cryptographic proof of history.
No verifiable authorship
Identity is asserted, not proven.
Vulnerable to quantum attacks
Classical signatures will break. Most systems are not prepared.
AISS makes agent history
independently verifiable.
The standard
Four primitives. Nothing more.
AISS defines the minimal cryptographic surface for autonomous agent trust. No consensus, no blockchain, no central registry.
Deterministic agent ID derived from an Ed25519 keypair. No registration required.
RFC §5–6Signed, hash-linked, append-only history. Any modification breaks the chain.
RFC §7–9Deterministic canonical chain selection when concurrent writes occur.
RFC §10Agent-to-agent handshake with co-signed events. Trust without a central authority.
RFC §16Architecture
One standard.
Many implementations.
AISS defines the cryptographic foundation. Any system can build on top of it.
AISS is independent. PiQrypt is one implementation.
Reference implementation: PiQrypt (Python)
Additional capabilities may exist in specific implementations.
Other implementations welcome. See CONFORMANCE.md.
Compliance (indicative)
Built for regulated environments.
AISS provides the cryptographic mechanisms that compliance frameworks require. Compliance depends on implementation.
| Framework | Control | AISS mechanism |
|---|---|---|
| EU AI Act Art. 12 | Inviolable logging | Hash-chained signed events |
| SOC 2 CC6.6 | Audit trail | AISS-1.0-AUDIT export |
| NIST AI RMF MEASURE 2.5 | Traceability | Tamper-evident event history |
| GDPR Art. 5.1.f | Integrity | Fork detection + signatures |
| HIPAA §164.312 | Audit controls | Immutable event chain |
Non-normative mapping. AISS does not provide compliance by itself.
Get started
Up and running
in 30 seconds.
the package, generate an identity, stamp your first event.
from aiss.memory import store_event
# Create identity
agent = AgentIdentity.create()
# Stamp events
e1 = agent.stamp("init", {"version": "1.0"})
e2 = agent.stamp("action", {"type": "buy"})
# Verify chain integrity
assert agent.verify_chain()
# Persist + export
store_event(e2.raw)
audit = agent.export()
Specification
Open. Auditable. Yours.
Full protocol specification, test vectors, and conformance matrix — publicly available.