The 2026 AI Quota Gray-Market Trust Crisis — A Sourced Timeline
Published by the MarketNow team (AliceLabs LLC). We build trust infrastructure for AI agents; our own audit evidence is public. This dossier only lists documented, linkable events.
If you buy Claude / Cursor / Codex access through resellers — shared seats, relay stations, budget subscriptions — 2026 was the year the market's trust problem became measurable. This is the timeline, with sources.
Timeline
2026-05-06 — A Chinese AI content creator documents being defrauded by two Claude relay resellers ("7折Claude" discount resellers and WorldClaw). Source: X post by 卡尔的AI沃茨
2026-05-07 — CHINAZ publishes a survival guide for the relay-station ecosystem, noting the three barriers (overseas phone, foreign cards, network restrictions) that create the gray market in the first place. Source
2026-05-12 — Sina Finance: "1 yuan for 2.85M tokens" — industry tests find malicious code injection, cloud credential theft and crypto theft inside relay stations. Source
2026-06-08 — Zhihu column (知乎专栏) investigation: relay stations accused of account bans, sellers vanishing ("跑路"), model quality degradation ("降智") and reselling user data. Source
2026-06-22 — China News Service / 中新经纬: a Shanghai relay-station operator was criminally detained for 37 days for illegally obtaining API models; publishes closure statement after release on bail. Source
2026-06-29 — TMTPost (钛媒体): "The price of buying Claude at 5% cost" — user reasoning traces logged and resold to data brokers, model trainers, and fraud profiling. Source
2026-08-05 → 08-16 — Linux.do accumulates scam-warning threads: "GPT 20x carpool seats diluted" (91 replies), "PRO20x carpool scam" with victims organizing rights-protection groups, home-broadband-sharing scam. Thread 1 · Thread 2
2026-08-18 — V2EX warning thread about a quota reseller: plans sold as "Composer/Auto unlimited" hit throttling mid-cycle with ~20% paid premium quota left; the thread's final demand: "the problem is no longer price — it's buying under one rule and getting another a week later." Source
Ongoing (Aug–Sep 2026) — Russia's payment crisis produces weekly "how to pay for Claude/ChatGPT" guides (Habr, vc.ru, RBC, DTF) as Russian BINs are systematically rejected; the gray-reseller market around them grows. Example (Habr) · Example (vc.ru)
The pattern
Every event above is a different failure mode — rule changes, exit scams, data resale, criminal detention, payment exclusion. But the root cause is one: what you pay for has no verifiable carrier. No signed record of the rules you bought. No third-party-verifiable uptime or quota accounting. Screenshots as the only evidence.
What verifiable would look like
- Every rule/quota change emits an Ed25519-signed receipt (non-repudiable, independently verifiable).
- Quota policies are versioned; your purchase pins the version.
- Verification uses canonical serialization (RFC 8785 JCS) and public endpoints — math, not self-reported dashboards.
We build exactly that for MCP servers and AI agents, and we published what happened when a public skeptic attacked it: 2 real bugs fixed within 24 hours, a symlink-escape gap disclosed with a working exploit and leak proof, byte-exact test vectors, deny logs (JSONL), and a reproducible harness — 25 PASS / 0 FAIL / 2 KNOWN-GAP.
- Evidence: https://github.com/alicelabs-llc/universal-trust-adapter
- Deny logs & harness: https://github.com/alicelabs-llc/universal-trust-adapter/tree/main/evidence/codepass-multichannel
- Verify a trust card (no registration): https://marketnow.site/verify
Also published in this series
- English: https://telegra.ph/You-Paid-an-AI-Reseller--Then-the-Rules-Changed-Mid-Cycle-09-08
- 中文: https://telegra.ph/你买的-AI-合租中转被改规则或跑路开发者自保清单附密码学验证思路-09-08
- Русский: https://telegra.ph/Oplatili-AI-podpisku-cherez-posrednika--a-pravila-izmenilis-posredi-cikla-09-08
- Español: https://telegra.ph/Pagaste-por-Claude-o-Cursor-a-un-revendedor-y-las-reglas-cambiaron-a-mitad-del-ciclo-09-08
2026-09-08 · MarketNow (AliceLabs LLC)