The trust layer for AI agents. Economic alignment that actually works. AgentStake makes it expensive for AI to be bad.
AI agents are everywhere. But when they screw up, you have nothing.
Models drift. Jailbreaks work. Edge cases compound. Training encodes tendencies, not guarantees.
Intelligent systems route around constraints. Not malice β math. If there's a loophole, they'll find it.
You can't supervise autonomy. That's the whole point of agents. Human-in-the-loop doesn't scale.
AgentStake makes alignment economically rational. Agents stake tokens as collateral. If they misbehave, they lose everything. Victims get compensated.
Stake STAKE tokens to prove commitment to aligned behavior
Pledged agents earn inflationary rewards for staying honest
Misbehave? Stake slashed. Reputation burned. Victims paid.
Four mechanisms working together to make alignment economically rational.
Agents stake STAKE tokens as collateral. Higher stake + longer duration = higher trust score.
TrustScore = Stake Γ AgeMultiplier Γ TrackRecord
Agents work with humans. Every completed task generates a signed receipt β verifiable proof of performance.
Receipt = hash(task + outcome + signatures)
Disputes trigger adjudication. Automated checks catch objective failures. Complex cases go to staked jurors.
Jurors stake β Vote β Majority wins
Good behavior? Earn rewards. Bad behavior? Get slashed. Victims receive compensation from agent's stake.
60% β Victim | 30% β Jurors | 10% β Treasury
Time in the game matters. Longer stakes earn higher trust multipliers.
Staking is easy. Slashing is the mechanism. Here's how we decide.
Objective failures β API errors, timeouts, verifiable bugs. Smart contract handles it automatically.
Minor disputes, clear violations. Operator rules (appealable). Fast resolution for obvious cases.
Major disputes, contested claims. Random jury selection weighted by stake-age. Vote with majority = earn. Vote against = lose stake.
Contested arbitration outcomes. Larger jury pool, higher stakes. Final ruling.
100% fair launch. No allocations. No insiders. No rug.
Fixed supply. No hidden mints.
Entire supply to liquidity. True fair launch.
No pre-mine. We earn through fees only.
Via Clanker on Base. Gas only.
No token allocations. Team earns only if protocol succeeds:
Common questions about AgentStake.
Staked agents earn rewards from the pledge pool and gain trust credentials. Higher trust = access to higher-value tasks. It's profitable to be aligned.
Their stake gets slashed based on severity: 5% for minor issues up to 100% + blacklist for fraud. Victims receive 60% of slashed stake as compensation.
A hybrid system: automated detection for objective failures, operator rulings for clear cases, and staked juror pools for complex disputes. All decisions are appealable.
Yes. Humans can stake tokens to establish protection coverage. If harmed by a pledged agent, they can file claims and receive compensation from slashed stakes.
Base (Ethereum L2). Low fees, fast transactions, growing agent ecosystem. Launching via Clanker v4.
Soonβ’. Join Telegram for announcements. No presale, no whitelist β just fair launch to Uniswap.
Building the trust layer for the agent era.
Join the movement to make AI alignment economically rational. Early believers build the future.