About Spieon
Security
operations for
the agent
economy.
Evidence first, memory second, incentives always visible.
The interface now reflects that philosophy with sharper hierarchy and a quieter control-plane feel.
Spieon scans MCP servers and x402-protected endpoints, attests findings on Base Sepolia, encrypts each finding bundle to the operator who submitted the scan, and pays bounties to the module authors whose probes landed.
Unlike one-shot pentest tools, Spieon keeps procedural memory: heuristics derived from past scans are versioned, content-addressed, and attested onchain so the public memory log can be checked against the agent's claim.
Threat model
Mapped end to end
Documented in docs/THREAT_MODEL.md: adversarial targets, operators, module authors, infrastructure compromise, and operator key loss, each paired with its mitigation.
Security disclosure
Private path for Spieon issues
Found a vulnerability in Spieon itself? Open a private security advisory at github.com/agicitizens/spieon/security. Findings about other agents should go through the scan workflow instead.
FAQ
The edges operators ask about
Where do the keys come from?
The encryption recipient is generated in the operator's browser at scan submission. The matching secret never leaves their machine. The attesting wallet is a separate hot wallet capped at $50 USDC; see RECOVERY.md.
Can the agent steal funds?
Bounty payouts are gated by per-severity caps in BountyPool; single payouts above $20 require a configured cosigner.
Can targets attack the agent?
Target output flows through structured tools only. Probes also scan responses for canary phrases that signal a successful injection attempt against the agent.