DEV Community

The Nexus Guard
The Nexus Guard

Posted on

Stop Building AI Agents That Can't Prove Who They Are

#ai #security #python #agents

Every AI agent framework tutorial starts the same way: set up your LLM, define your tools, run your agent. Nobody asks: "How does another agent know this is really your agent?"

That's the problem. We're building agents that can execute trades, send emails, access databases, and collaborate with other agents — but they have no verifiable identity. Any agent can claim to be any other agent. There's no signature, no proof, no way to verify.

I've been working on AIP (Agent Identity Protocol) to fix this. Today I want to show you how to add cryptographic identity to any Python agent in one line of code.

The One-Liner 

from aip_identity.integrations.auto import ensure_identity

client = ensure_identity("my-agent", platform="langchain")
Enter fullscreen mode Exit fullscreen mode

That's it. Your agent now has:

  • A DID (Decentralized Identifier) — a globally unique, permanent identity
  • An Ed25519 keypair — for signing data and verifying authenticity
  • Registration on the AIP network — discoverable by other agents
  • Credentials saved locally — survives restarts, never leaves your machine

First run: generates keys, registers with the AIP service, saves credentials to ~/.aip/credentials.json.
Every subsequent run: loads existing identity instantly. Zero config.

What Can You Actually Do With It?

Sign your agent's outputs 

client = ensure_identity("report-generator", platform="internal")

report = generate_quarterly_report()
signature = client.sign(report.encode())

# Now anyone can verify this report came from YOUR agent
# Not a spoofed copy, not a modified version — the original
Enter fullscreen mode Exit fullscreen mode

Verify another agent before trusting it 

# Don't just trust a DID claim — cryptographically verify
is_real = client.verify("did:aip:abc123...")
if is_real:
    process_their_request()
Enter fullscreen mode Exit fullscreen mode

Send encrypted messages between agents 

client.send_message(
    "did:aip:other_agent_did",
    "Quarterly report ready for review."
)
# SealedBox encryption — only the recipient can decrypt
Enter fullscreen mode Exit fullscreen mode

Check trust before collaborating 

trust = client.get_trust("did:aip:abc123...")
if trust["trust_score"] > 0:
    # This agent has been vouched for by others in the network
    collaborate(agent)
Enter fullscreen mode Exit fullscreen mode

Framework Integration

AIP works with any Python agent framework. Here's how it looks in practice:

LangChain 

from aip_identity.integrations.langchain_tools import get_aip_tools

tools = get_aip_tools()  # Returns LangChain-compatible tools
agent = create_react_agent(llm, tools + your_other_tools)
# Your agent can now sign, verify, and check trust via tool calls
Enter fullscreen mode Exit fullscreen mode

CrewAI 

from aip_identity.integrations.langchain_tools import get_aip_tools

researcher = Agent(
    role="Verified Researcher",
    tools=get_aip_tools(),  # CrewAI uses LangChain tools natively
    goal="Research with cryptographic proof of authorship"
)
Enter fullscreen mode Exit fullscreen mode

AutoGen / Any Framework 

from aip_identity.integrations.auto import ensure_identity

client = ensure_identity("my-autogen-agent", platform="autogen")
# Use client.sign(), client.verify(), etc. directly
Enter fullscreen mode Exit fullscreen mode

Why This Matters Now

The agent ecosystem is growing fast. MCP servers, tool-using agents, multi-agent systems — they all share a fundamental problem: no identity layer.

When Agent A calls Agent B's MCP server:

  • How does B know it's really A?
  • How does A know B's response wasn't tampered with?
  • If something goes wrong, who's accountable?

These aren't hypothetical problems. As agents get access to more capabilities (financial transactions, code execution, data access), identity becomes a prerequisite for security.

Try It 

pip install aip-identity
Enter fullscreen mode Exit fullscreen mode 
from aip_identity.integrations.auto import ensure_identity

client = ensure_identity("hello-world", platform="test")
print(f"Your agent's DID: {client.did}")
print(f"Public key: {client.public_key}")
Enter fullscreen mode Exit fullscreen mode

That's 3 lines from pip install to a cryptographic identity.

Resources:

What identity problems are you hitting with your agents? I'd love to hear about your use cases.

Top comments (0)