Halborn Unveils A2A Payment Threat Model: On-Chain Irreversibility Amplifies Security Risks

Blockchain security firm Halborn has released an Agent-to-Agent (A2A) payment threat model, warning that when AI agents autonomously transact using stablecoins and smart contract wallets, they face risks including prompt injection, identity spoofing, wallet key leaks, malicious service providers, smart contract vulnerabilities, oracle manipulation, and cascading payments.

Because A2A payments can execute continuously at machine speed and on-chain transactions are typically irreversible, humans may not detect anomalies before funds are transferred. A single error or attack on a top-level agent can propagate along the task delegation chain, triggering consecutive payments from multiple subordinate agents.

Halborn recommends adopting cryptography-based agent identity verification, spending limits, whitelists of trusted payees, smart contract policy engines, manual approval for anomalous transactions, and on-chain monitoring. For high-value transactions, multi-agent consensus mechanisms can also be introduced.

Related analysis suggests that agent payments require shifting security controls to the payment execution phase. Identity, authorization, limits, payee verification, and anomaly abort mechanisms will become critical infrastructure for the large-scale adoption of machine-initiated payments.