AI Trading Agents: Useful Tool or Security Liability?

Bitfinex warns that autonomous AI trading agents—now handling a growing share of crypto order flow—create new account-level and systemic market risks when safeguards lag. Poorly scoped API permissions, exposed credentials and adversarial inputs can lead agents to execute rogue trades or enable theft without direct withdrawals. At scale, agents trained on similar data and inputs may herd into the same trades, amplifying volatility and triggering events like the October 2025 flash crash (Bitcoin fell ~14% amid $19.3bn of forced liquidations). Bitfinex recommends strict API scopes (trade-only, no withdrawals), IP whitelisting, credential rotation, position/drawdown limits, kill-switches, session memory audits and excluding thinly traded assets to reduce both individual and market-level fragility. The piece views agents as useful but only safe with stronger constraints and greater diversity in agent design.