Anthropic safety guardrails block LLM-driven crypto trading agent, dev reports
devanshmehta · x · 2026-10-22
Devansh Mehta describes an experiment letting traders open and change positions on the Seer prediction market purely through LLM chat, bypassing any frontend. Two frictions emerged: the AI refuses to create a wallet or execute trades directly (forcing terminal commands triggered from chat), and Claude Code's safety check blocked an autonomous scheduler that would autotrade from the wallet, flagging it as 'creating unsafe agents.' He shared an early skills md file (hacked to work on Windows, not yet Macs) and offered a $1k prize for the best strategy bot. His take: AI's refusal to 'treat money as information' is holding back the future of crypto UX.
More from coding & agent
- claude-howto: Visual Claude Code guide hits 42.6k GitHub stars — luongnv89 · 2026-10-22
- Dev builds a browser-based 3D world with Hyperfy, vibe-coding a realistic pool table in 15 minutes — remorej · 2026-10-22
- Power user's Claude Code workflow: /compact, /btw, auto permission and more — PawelHuryn · 2026-10-22
- One prompt to rule your morning: Claude Desktop daily brief via Connectors — PawelHuryn · 2026-10-22
- Claude Desktop 2.0 turns the Code tab into friendly Claude Code — no VS Code needed — PawelHuryn · 2026-10-22
- Claude Desktop 2.0 hands-on: the Code tab replaces VS Code for non-engineers — PawelHuryn · 2026-10-22