GCVE-110-VULN-PYPI-2026-000636

If you run aimarket-platon on a network that hosts or containers you do not control can also reach (a LAN, a shared Docker network, a Kubernetes namespace), any of them can reset the per-client rate limit on the paid ask completion (POST /api/ask and the platon.ask@v1 capability) with every request and spend your model budget past the 30-a-minute ceiling. The server binds 0.0.0.0 by default and keys the limit on X-Real-IP, or the right-most X-Forwarded-For entry. 0.1.5 reads those headers only when the TCP peer is a loopback, private or link-local address (main._from_local_proxy, main._key_from), on the assumption that such a peer is your reverse proxy. A neighbour on a private network is such a peer, so it still names its own key. In 0.1.0 through 0.1.3 a caller on any network could (GCVE-110-VULN-PYPI-2026-000615). No release limits header trust to your proxy. Affects 0.1.0 through 0.1.5.