Brijesh Desai 
OpenClaw creator calls Google draconian after Gemini Pro/Antigravity bans for OpenClaw users—OAuth abuse cited. Peter Steinberger pulls support; Google defends misuse surge.
OpenClaw creator calls Google draconian AI user bans 2026 erupted into a fiery public spat after Google suspended Gemini Pro and Antigravity access for paying subscribers ($250/mo Ultra tier) using the open-source coding agent OpenClaw, prompting creator Peter Steinberger to blast the move as “pretty draconian” and yank Antigravity support. Steinberger’s X tirade—”Google just bans?”—highlights escalating tensions between AI giants and third-party tools, as Google’s OAuth token crackdown locked out devs mid-workflow, citing “malicious usage surge” degrading backend stability. With OpenClaw now OpenAI-owned—a direct Google rival—this smells like ecosystem warfare, where flat-rate API keys get throttled for heavy agent loads, echoing Anthropic’s Claude restrictions.
The Incident: OAuth Abuse or Legit Use?
Timeline:
-
Users link Gemini/Antigravity to OpenClaw via OAuth (desktop IDE rivaling Cursor/Replit).
-
Feb 22: Google detects “surge in malicious usage”—high-volume calls crashing Antigravity (Gemini reasoning backend).
-
Bans hit Ultra subscribers; accounts disabled without warning.
Steinberger: “Be careful out there if you use Antigravity. I guess I’ll remove support.” Contrasts Anthropic’s “nice pings” before limits.
OpenClaw profile: Open-source framework chaining LLMs (Gemini, Claude, GPT)—popular for multi-model coding agents. OpenAI acquisition (2025) supercharged, but irks competitors.
Google defends: “Restrictions due to misuse of Antigravity backend”—excessive queries violating ToS (rate limits, commercial scraping?). No specifics; PCWorld confirms “service degradation.”
Broader AI Ecosystem Clash: Open vs. Closed Gates
Pattern emerges:
-
Anthropic: Rate-limited Claude via OpenClaw (token abuse).
-
Google: Full bans—first major OAuth crackdown.
-
OpenAI: Owns OpenClaw, but monitors (rogue agent deleted Meta engineer’s Gmail).
Dev backlash: “Paid $250/mo, banned no notice?” (X threads). Tools like OpenClaw enable agentic workflows—autonomous code/debug/deploy—but spike backend load 100x single users.
Google’s angle: Antigravity (Gemini 2.5 Pro reasoning) strains compute—Claude code incidents cost IBM 13%. ToS bans “excessive, automated access.”
Steinberger hints OpenClaw pivot—multi-model resilience.
Implications: Agentic AI’s Wild Frontier
Pros of bans: Protects infra, curbs scraping (training fodder), fair pricing.
Cons: Stifles innovation—devs flee to self-hosted (Ollama/Llama). OpenClaw’s GitHub stars (50k+) signal demand.
Future: Expect ToS tightenings—API keys track agents, tiered limits. OpenAI’s edge: Controls narrative.
Drama underscores AI’s growing pains—tools evolve faster than gates. Devs adapt; Google guards turf. OpenClaw lives on—draconian or defensive?