Google restricting Google AI Pro/Ultra subscribers for using OpenClaw

Original Article ↗ Hacker News Discussion ↗

What actually happened (as discussed)

  • Users used Google AI Pro/Ultra “Antigravity” OAuth tokens with OpenClaw/OpenCode instead of paying for the official Gemini API.
  • These integrations impersonate the Antigravity client (reusing its OAuth client ID/flow) and then call private “Cloud Code” endpoints directly.
  • Google responded by suspending access to Antigravity and Gemini CLI for those accounts; other Google services (Gmail, Photos, etc.) appear unaffected, but this is a point of confusion and fear.
  • Similar reports exist for other tools (OpenCode, Gemini-auth plugins), and Anthropic has taken parallel measures with Claude Code tokens.

Is this a legitimate ToS violation?

  • One group says yes: this is clearly using a private, subsidized internal API outside its intended client, akin to scraping a Netflix app or abusing an all‑you‑can‑eat buffet. If you want programmable access, buy API credits.
  • Others argue the UX made this look “official enough” (Google-branded OAuth dialog), and if Google didn’t want this, they should have scoped OAuth properly or rate-limited instead of retroactively nuking access.
  • Strong criticism targets the zero‑tolerance, no‑warning bans and continued billing of $200–$250/month for unusable plans.

Economics, subsidies, and anti‑trust concerns

  • Many note that subscription tiers give vastly more tokens than equivalent API spend; users are “subsidized” and some were burning thousands of dollars of compute for $200.
  • Defenders frame this as normal loss-leading, prompt‑caching, and data-collection economics; critics call it predatory cross‑subsidization designed to kill competition, then rug‑pull.
  • There’s debate over whether inference is still heavily loss-making or already profitable at API prices; no consensus in the thread.

Trust, lock‑in, and ecosystem effects

  • The bans reinforce long‑standing fear of losing a 10–20‑year Google identity over one product’s ToS, prompting calls to de‑Google, self‑host email, and regularly export data.
  • Some see this (and Anthropic’s moves) as pushing developers toward open‑weight or Chinese models (GLM, Kimi, MiniMax) and local LLMs, despite their higher hardware requirements.
  • OpenClaw is viewed as existentially threatening because it makes model providers interchangeable; harsh enforcement is interpreted as an attempt to keep users inside first‑party toolchains.

Technical and policy critiques / proposed alternatives

  • Commenters note Google could:
    • Enforce documented quotas and throttling on Antigravity,
    • Issue clear warnings and temporary suspensions,
    • Redirect heavy “agent” usage to paid API plans,
    • Or design a separate, constrained “subscription-only” API.
  • Later, a Google employee cited “massive malicious usage” degrading service and promised a path back for unaware users, but earlier support emails explicitly said suspensions were irreversible, worsening the perception of chaotic, user‑hostile enforcement.