AI agents worry crypto firms
Security teams at Bitget and SlowMist warned that autonomous AI trading agents — if weaponized via prompt injection or malicious plugins — pose systemic crypto risks that could affect BTC and broader markets. Separately, demos are exploring Bitcoin + Lightning as a counterparty‑free settlement layer for machine agents, intensifying the debate. (x.com) (x.com)
Bitget and SlowMist published a joint AI Agent Security Report in mid‑March 2026 that frames automated trading as entering an “agentic” phase and maps seven layers of threat across development, orchestration, plugins and on‑chain execution. (pressreleasehub.pa.media) SlowMist’s analysis flagged mass‑uploaded malicious plugins in OpenClaw’s ClawHub — 341 flagged out of 2,857 scanned, about 12% — and other monitors reported roughly 10% of plugins in agent marketplaces showed two‑stage malware behavior. (coinalertnews.com) The report and independent demos call out indirect prompt‑injection and supply‑chain plugin poisoning as practical, high‑impact attack chains, with security researchers showing 0‑click exfiltration and instruction‑injection that can override agent decision logic. (cybersecuritynews.com) To reduce risk, SlowMist and Bitget proposed concrete controls — minimum‑privilege APIs, sub‑account isolation, human‑in‑the‑loop confirmation signatures and a five‑tier security governance framework that includes continuous monitoring and on‑chain risk analysis. (chaincatcher.com) On the settlement side, Lightning Labs released open‑source LN agent tools in February 2026 that let autonomous agents run Lightning nodes, create invoices, and pay without identity or API keys, implementing machine‑to‑machine payment flows. (theblock.co) The space is fragmenting into competing rails: Coinbase and others are advancing x402/Agentic Wallet concepts while Stripe and payment teams are piloting USDC machine payments — a fast‑emerging protocol race over which settlement layer agents will use. (bitcoinmagazine.com)
