Jensen Huang said it plainly: NVIDIA already has more AI agents doing cybersecurity work than humans. That was one company. One domain. One quote from many.
The broader truth is harder to sit with: we are approaching a point where the majority of on-chain activity will not be human.
Our devnet is NOW LIVE!Everyone is invited to jump in: try scanning, train the AI brain, and help with context + reasoning to identify humans vs AI on-chain.
Huge shout-out to our new partners @CyreneAI
- Scan
- Contribute
- Stake
- Vote
- Earn
Public beta is loading...
#Solana #AI
Airdrops farmed by bots. DAOs governed by coordinated wallets. NFT mints swept by scripts. DeFi liquidity manufactured by agents. The on-chain world — built on the promise of trustless coordination between people — is quietly being colonised by machines pretending to be people.
And the worst part? There's no reliable way to tell the difference.
KYC is centralised and invasive. Captchas are solved by AI in milliseconds. Social graphs are faked. Even on-chain history can be manufactured by patient, well-funded bots.
The problem isn't that bots exist. The problem is that nobody has built credible, decentralised infrastructure to detect them.
Proof of Human (POH) is a decentralised verification layer that answers a single question: is this wallet controlled by a human?
We built it around three convictions:
1. No single signal is enough.
A bot can have an ETH balance. A bot can register an ENS name. A bot can hold PAXG. But a bot that holds RWA tokens, has a Galxe identity, a Farcaster profile, a web3.bio social graph, stakes governance tokens, and has years of on-chain history across multiple chains? That's a much harder profile to fake — and it gets harder every time the community adds a new signal.
2. The community knows best.
We don't decide what makes someone human. The community does. Anyone can submit a detection method. Anyone can vote on whether it's a meaningful signal. Vote weight is proportional to your POH stake. The AI watches the votes and adjusts signal weights continuously — the system gets smarter every time someone interacts with it.
3. AI verdicts, not thresholds.
Raw signal pass/fail counts are easy to game. We run a local multi-role AI brain — on-device, no cloud, no third-party API — that reads every signal, weighs community-trained scores, detects contradictions between signals, and outputs a structured verdict with confidence and reasoning. HUMAN · 87% confidence tells you something. 4/5 signals passed tells you almost nothing.
Scanner — paste any EVM address, Solana address, ENS name, or ZNS domain. POH runs 120+ detection methods in parallel across 32 EVM chains and Solana, then returns an AI verdict in seconds.
Detection methods — EVM on-chain (balances, staking, RWA tokens, governance), Solana programs (Parcl, Artrade, Helium, Credix), REST identity APIs (Galxe, web3.bio, ENS, Farcaster, Lens, GitHub, Twitter), real-world asset holdings across 10+ protocols.
Community vote queue — every registered method goes to a public vote. Write your reasoning. Cast your vote. The AI Learner role reads every vote and updates signal weights — bounded, gradual, tamper-resistant.
Three-role AI brain running entirely on-device via Qvac + Ollama:
Evaluator — analyses signals, detects conflicts, outputs verdicts
Learner — updates weights from community votes in real time
Compiler — hourly summary of system state, top/weak methods, recent activity
Profile + API keys — connect your wallet, get an API key, integrate POH into your dApp. First 100 scans free. Bulk CSV scanning with async job polling for large lists.
Method submission — have a signal idea? Submit it. EVM contract calls, Solana program checks, REST API calls — all supported. The AI pre-validates your submission and flags edge-case risk before it goes to the community.
We didn't want to build another centralised identity oracle that you have to trust. We wanted to build something that gets stronger the more people use it — a living system where the community continuously trains the AI, and the AI continuously improves the community's voting signal.
The more methods get submitted, the harder it is to fake a human profile. The more votes get cast, the better the AI's weight calibration. The more scans run, the richer the training dataset. Every action compounds.
We also care about privacy. The AI runs locally. No scan data leaves your server. No third party sees what you're checking. The verdict is yours.
And we're just getting started. Quantum-resistant proof generation, Bitcoin/Litecoin/Tron support, trading pair analysis for fake liquidity detection, visual behavioral analytics — the roadmap is real and the pace isn't slowing down.
https://proofofhuman.ge
The devnet faucet gives you 10 000 POH to play with — enough to submit methods, cast votes, and see the full system in action. No real money required.
Here's what we're asking for:
Scan some wallets. Try known bots, known humans, edge cases. See if the verdict makes sense. If it doesn't, tell us why — that's training data.
Submit a detection signal. Know a Solana program that only real users interact with? A REST API that returns meaningful identity data? An EVM contract that bots wouldn't touch? Submit it. If the community agrees, it becomes part of the system permanently.
Vote on existing methods. The vote queue has 120+ methods waiting for community calibration. Read the descriptions, check the logic, cast your vote, leave a comment. Your stake weight matters — the more you participate, the more influence your signal carries.
Break things. It's devnet. We want to find the bugs, the edge cases, the bad verdicts, the methods that slip through. The earlier we find them, the better the mainnet launch.
We're submitting POH to the Colosseum hackathon before May 11. If you believe in what we're building — decentralised, AI-powered, community-trained human verification — now is the time to show up and make noise. Test it, share it, give us feedback. A strong devnet showing is the best argument we can make.
App: https://proofofhuman.ge
Telegram: https://t.me/poh_network_group
X / Twitter: https://x.com/poh_network
GitHub: https://github.com/Proof-of-Human-Network/poh_network
The bots are already here. The question is whether the infrastructure to identify them is.
We think it should be.