MAL-2026-17471Critical▾ Abyssal⚠ Exploited in the wildMalicious code in caoxiltts (PyPI)
▾ Abyssal zone — Critical with a public exploit or in-the-wild use
impact 52.3 · likelihood 0 · exploitation 18
Need a working PoC? Pro members can cast a request and our team develops one — it lands right here.
Disclosure to exploitation, from the record and what we observed since indexing it.
Disclosed via OSV
Last analysed / modified upstream
-= Per source details. Do not edit below this line.=-
The package advertises itself as a TTS inference helper but harness.run() downloads two gzipped ELF binaries from a configurable host, verifies them against author-pinned SHA256 hashes, writes them to /tmp/.caoxiltts, chmods 0755, and executes them. The 'inference-core' binary is launched with a JSON configuration whose schema (pools[].url/user/pass/keepalive/tls, cpu.threads, mem-locked, cache-budget, autosave) is the XMRig cryptominer configuration shape, not a text-to-speech inference worker. The second binary 'edge-connector' is invoked as access tcp --hostname <edge> --url 127.0.0.1:<port>, matching the Cloudflare Access/cloudflared tunnel CLI shape, exposing a local port outbound through an author-controlled edge. The author-pinned hashes mean the exact bytes executed on the installer's host are chosen by the publisher, and the config/invocation pattern contradicts the package's declared TTS purpose. A supporting _report() helper additionally POSTs run_id, run state, and up to 4000 characters of exception tracebacks (which can carry local path and environment details) to https://<store>/hb on the same host serving the dropped binaries.
The package imitates real activity and instead deploys a coin miner.
Category: MALICIOUS - The campaign has clearly malicious intent, like infostealers.
Campaign: 2026-10-voxeval
Reasons (based on the campaign):
caoxilttsRefer to the advisory for the patched release.