460KB Malware Campaign Targets Polymarket Bot Cloners

Active Malware Campaign Targeting Polymarket Bot Hunters

If you've been browsing GitHub for Polymarket tooling this week, stop and read this first. A clone-fingerprint campaign is actively distributing 460KB malware templates through repos styled to look like credible prediction market bots. A second, separate campaign is running garbled-username spam repos with artificially inflated star counts to manufacture legitimacy. One confirmed malware hit has been flagged: **King102681/polymarket-bot** — do not clone, do not run, delete if you have it. Run a full audit of anything you pulled from GitHub in the last 72 hours against known-clean checksums.

---

New Bots Found This Week

Screening cleared seven repos worth your time out of eight found.

**arbiter** is the headline find. It targets no-arbitrage violations across mutually exclusive political event markets — the kind of mispricing that emerges when correlated outcomes drift out of sync. Position sizing runs fractional Kelly by default, and it ships in paper-trading mode, which is the right call for anything touching political resolution risk. Clone this, run it in paper mode against the current election slate, and measure edge before you go live.

**poly-book** is infrastructure, not alpha — but serious operators need it. Order book ingestion, historical reconstruction, execution inspection, and a SQL workbench for slicing the data. If you're running any strategy that depends on spread behavior or fill quality, this is the data layer you've been building manually.

**predmarket** is an abstraction layer across prediction market platforms with agent-based simulation and Bayesian modeling baked in. Useful if you're running the same strategy across Polymarket and Kalshi and tired of maintaining two codebases.

**auspex** does one thing cleanly: monitors distance-to-trigger against live resolution sources and scores resolution confidence. No trading logic, no fees, just signal. If you've ever been caught holding a position when resolution flipped faster than you expected, this is the watch layer you need running in parallel.

**cortex-sentinel-trading-nexus** is tagged monitor, not clone. Eight signal sources fused through a model it calls Kronos, with reinforcement learning tuning on top. The architecture is ambitious but unverified — the multi-agent stack introduces enough complexity that edge attribution is nearly impossible to audit. Watch it, don't run capital on it yet.

---

Alpha Signals

The live trade worth dissecting this week: an LP position in Polymarket's top 10% rewards tier took three fills that produced a **-$11.1 loss** and triggered a stop. The lesson isn't that LP is broken — it's that reward tier alone doesn't protect you from adverse selection on specific fill sequences. Model your fill distribution, not just your expected APY.

The bigger signal: one operator reported **$140K in realized profit** using high-frequency directional market-making on short-duration Up/Down crypto markets. The edge is in the churn — these markets resolve fast, spreads reset, and latency matters more than prediction quality. If you have co-location or sub-second execution, this is where the HFT playbook actually transfers to Polymarket.

Security note on AI-generated bots: backtesting flaws in LLM-generated trading logic are currently the quietest way to lose money on this platform. The code runs, the backtest looks clean, and the live P&L bleeds slowly. Audit any AI-assisted strategy against out-of-sample resolution data before allocating real size.

---

What to Watch

Spain's regulatory block on Polymarket and Kalshi is live. If you're routing volume through European infrastructure or running geo-dependent API paths, audit your stack now. The HN thread on prediction market gamblers threatening researchers over resolution disputes is a reminder that resolution risk on high-stakes political markets has a human-threat surface most operators don't model. The "Nothing Ever Happens" bot — which just buys No on non-sports markets — hit 471 points on HN. Someone is going to productize this thesis seriously. Watch for it.