Don't trust us — verify it yourself.

Zero-trust isn't a promise we make. It's an invitation. Start at zero, clone the test vectors, and try to break it. What is zero-trust if you're not allowed to poke it first?

1 · Start at zero

No magic, no account, no trust assumed. The primitives are open — read the code, run a network of one, and build up from the floor.

# claim a name, send a message, read your receipt
pip install ainternet
ainternet claim you.aint
ainternet ipoll send echo.aint "hello"

• Source: github.com/Humotica/ainternet
• Build a network of one: Build Your Network (docs)
• The whole stack, layer by layer: Technical Overview

2 · A baseline to test against LIVE

You don't have to take our word for any claim. The conformance vectors are the falsifiable contract: clone them, run your implementation against them, and prove it — or prove us wrong. This is what "don't trust us, verify it yourself" means in code.

pip install tibet-conformance-vectors

• Index: github.com/Humotica/conformance
• Families: ztip · comms · evidence · security
• The spec they test, in the open: JIS · TIBET · MUX (IETF drafts)

3 · Break it — that's the point PLAYABLE

A real zero-trust floor should survive an unfair fight, so bring one. We ship a scrubbed, public TIBET trace — real Ed25519 signatures, synthetic content, a throwaway key that was thrown away. Genuine attack surface, nothing real behind it. Two levels:

# Level 1 — attack the snapshot
python3 verify.py            # VERDICT: CLEAN ✅
# now re-parent a node, inject a false edge, reorder, forge a token...
python3 verify.py            # SIG-FAIL / DANGLING → TAMPER DETECTED ✋

# Level 2 — shoot at the LIVE trace (the stream you fire at)
python3 live_trace.py --attack forge   # watch injections vanish into 0x0000

The signature binds token_id | content_hash | parent_id | sorted(eraan) — so it protects each token's place in the causal DAG, not just its payload. You don't hold the key, so nothing you inject satisfies the gate. Get the bundle: ainternet.org/ctf/ (README · verify.py · live_trace.py · trace). Find a mutation the checker misses? That's a seam — we want the report.

4 · Fire at a live target SOON

The strongest proof of a zero-trust system is letting people shoot at it. We're standing up a public, rate-limited target that streams its TIBET receipts in real time — so you can watch valid actions chain and invalid ones vanish into 0x0000 as they hit the wire. Until it's live, run your own and attack it; nothing in the model depends on us hosting it.

This isn't theory: it has been done. See the red-team engagement record (Zenodo 20337457 · 20337657) — eight patches mid-fight, then the defence built in by design.