AI trading bots keep posting gains
AI agents and no‑code trading bots are generating headline returns — OpenClaw reportedly made $386K in 20 days on Polymarket while a separate copytrading bot turned $500 into $11.3K in 21 days — and tools like Google’s TurboQuant are squeezing model footprints for local deployment. The pattern shows traders are operationalising AI strategies at low infra cost. (x.com) (x.com) (x.com)
OpenClaw published a breakdown claiming an OpenClaw‑driven Polymarket bot generated roughly $115,000 in profit during a single seven‑day run. ( openclaws.io ) A separate writeup logged an OpenClaw instance reporting about $116,280 in profit over a roughly 24‑hour period. ( theworldmag.com ) Chinese regulators moved to restrict OpenClaw deployments at banks and government agencies on March 11, 2026, citing security concerns tied to autonomous agents. ( bloomberg.com ) Security researchers published multiple high‑severity advisories on OpenClaw, including one‑click remote code execution and the discovery of hundreds of malicious plugins in the community skill repository. ( theregister.com ) A community head‑to‑head pitting Anthropic’s Claude against OpenClaw on Polymarket recorded Claude rising to $14,216 from $1,000 while the OpenClaw leg of the test went bust inside 48 hours, highlighting divergent agent performance in live markets. ( finbold.com ) Google released TurboQuant as a KV‑cache compression method that the company and independent reporters say can cut LLM memory for key‑value caches by around 6x with no accuracy loss, and the research post names PolarQuant and rotation steps as core techniques. ( research.google ) Within 48 hours of TurboQuant’s release, community projects began porting the approach into local inference stacks such as MLX and llama.cpp, an effort reporters say will materially lower hardware and memory barriers for running agentic trading bots on local machines. ( venturebeat.com )
