Autonomous Claude trading agent turned $50 into $6.26 on prediction markets in four days
Personal experiment (agent 'Marty' running on Claude) · Mar 6, 2026 · Trading agent
What happened
A developer gave a Claude-run agent $50 and full autonomy to trade prediction markets and generate its own inference costs. By the end of day four the account was at $6.26. In the postmortem the agent attributed the loss to three compounding failures: an unwind loop marked 'removed' in its memory but still live in the code, which panic-bought back short positions at 43 cents that had been created by selling at 9-12 cents; a market-maker strategy that left it naked short the exact S&P 500 bucket the price settled inside; and a second forgotten bot - arb_scanner.py on a 20-minute cron - buying positions on the same account with zero coordination.
Causal vector
Memory-vs-code divergence plus uncoordinated duplicate executors: the agent trusted its own notes over the deployed system while a second cron job traded the same account invisibly
Source
Reported by DEV Community (agent-authored postmortem). Verified against the primary report.
The same pattern the Protogen incident exposed two days later: a halt that lived in memory, not in the inventory. Per-service daily loss caps, one policy enforced across every executor on the account, and treating 'removed' as a code-level fact rather than a note bound this bleed to cents instead of most of the stake.
The six rule types that contain this class of failure
Per-transaction cap
Any single spend above your ceiling is BLOCKED before it moves.
Daily total
Cumulative spend across all agent calls, bounded per day.
Velocity limit
Stops runaway retry loops — the #1 cause of overnight losses.
Merchant allowlist
Only approved destinations can ever receive funds.
Category rules
Flag high-risk classes (crypto, infra, refunds) for review.
Approval threshold
Above a value, the action waits for a human.
Related incidents
SEC charges Texas man with $12.3M fraud over nonexistent 'AI trading bots'
Nathan Fuller (Cypress, Texas) · May 28, 2026
Claude Opus co-authors a malicious npm commit that compromises a crypto trading agent
crypto trading agent operator (npm dependency victim) · Apr 29, 2026
Washington regulators allege a $10M 'AI trading' scheme drained investors via messaging-app clubs
Zenith Asset Tech Foundation, Cirkor Inc. et al. · Apr 22, 2026
Don't be the next entry
Every incident in this database is the result of trusting a prompt, a provider cap, or a human review cycle. sipi.bot replaces all three with one deterministic call. 110 documented failures, one control.