Google Permanently Bans AI Pro Users for Accessing Gemini via OpenClaw, Continues Charging $250/Month
A Hacker News post garnering 140 points and 107 comments details how Google terminated Google AI Pro and Ultra accounts without warning after users accessed Gemini through OpenClaw, a third-party client. The incident surfaces deeper issues around prompt caching, subscription economics, and how AI providers enforce terms of service.
A post on Hacker News on February 23, 2026 drew 140 points and 107 comments after a user reported that Google had permanently terminated their Google AI Ultra account — without warning, and without stopping the $250/month billing — following use of Gemini through OpenClaw, a third-party client.
What Google Support Said
According to the affected user, Google’s support response was unambiguous: using OpenClaw to access Antigravity (Google) servers for non-Antigravity products violates the terms of service. The policy, the support agent stated, is zero tolerance, and the ban is non-reversible.
The specific detail that generated the most discussion on HN was not the ban itself, but the fact that Google continued charging $250 per month after locking the user out of their account.
The Technical Problem: Prompt Cache Destruction
Several HN commenters identified a technical dimension to the third-party client issue that goes beyond terms of service enforcement.
Prompt caching is a mechanism where AI API providers cache the leading portion of a context window — typically a static system prompt — so that subsequent requests can reuse it without reprocessing. When it works correctly, cache hit rates above 90% are achievable, reducing effective token costs by a factor of 10 or more.
OpenClaw, in at least some configurations, inserts the current timestamp (formatted as hh:mm:ss) at the very beginning of the context window on every request. Because the first bytes of the context change with every call, the cache is invalidated on every request — driving the hit rate to near zero.
Claude Code’s official developers have stated publicly that the client is “carefully designed to maximize prompt caching.” The design principle is straightforward: static content goes at the top of the context window; dynamic content goes at the bottom. Many third-party clients implement context construction without knowledge of this constraint.
The Economics: Subscriptions Were Not Designed for Agents
The broader issue underlying the ban is a growing mismatch between how AI subscription products are priced and how developer-heavy users actually consume them.
| Usage scenario | Monthly cost |
|---|---|
| Claude Pro / Google AI Ultra subscription | ~$200/month |
| Equivalent API usage (heavy developer workload) | $1,600+/month |
Subscription pricing assumes usage patterns typical of a human user interacting with the provider’s own client — browsing conversations, drafting documents, answering questions. AI coding agents like Claude Code or Gemini CLI can consume orders of magnitude more tokens per session than those baseline assumptions.
The ccusage CLI, which gained renewed attention following this incident, visualizes exactly this gap: it records Claude Code’s actual token consumption and calculates what the equivalent API costs would have been. For heavy users, the numbers frequently exceed subscription pricing by a wide margin.
Takeaways for Developers
Three practical conclusions emerge from this incident.
Use pay-as-you-go API keys for agentic workloads. Subscription products are designed for human-paced, client-mediated usage. Routing agent traffic through subscription accounts puts you in a gray area that providers — Google especially — are willing to terminate over.
Design for prompt cache efficiency. If you are building a client or wrapper that calls AI APIs, avoid placing dynamic content (timestamps, session IDs, request-specific metadata) at the beginning of the context window. Treat the cache prefix as immutable across requests.
Google’s zero-tolerance posture is not universal, but it is real. Other providers have handled similar situations with warnings or throttling before termination. Google’s policy, as stated in this case, skips those steps. Whether that approach is applied consistently remains to be seen, but developers should treat it as the baseline assumption.
Third-party clients exist in a structurally ambiguous position relative to AI providers. Unlike web browsers, which access services through documented, public HTTP interfaces, these clients are often routing requests through infrastructure designed for the provider’s own products. Where exactly that line sits — and how strictly providers choose to enforce it — is a question that this incident puts clearly on the table.
関連記事
GoogleがOpenClaw経由のGemini利用ユーザーのアカウントを永久停止——月額$250請求継続のまま
2026年2月23日、Hacker Newsで140pt/107コメントを集めたレポートによると、GoogleはOpenClaw(サードパーティクライアント)経由でGeminiを使用していたGoogle AI Pro/Ultraユーザーを予告なしに永久停止した。技術的・経済的背景を整理する。
Claude Code v2.1.47 リリース - Windows対応強化・メモリ改善・40件超のバグ修正
Anthropicは2026年2月19日、Claude Code v2.1.47をリリースした。Windows環境の描画バグ修正、長時間セッションのメモリ改善、起動速度の向上(約500ms短縮)、コンテキスト圧縮後のプランモード保持など40件以上の修正が含まれる。
Claude Code v2.1.49 リリース - バックグラウンドエージェント制御改善・起動高速化・Sonnet 4.6の1Mコンテキスト対応
Anthropicは2026年2月20日、Claude Code v2.1.49をリリースした。バックグラウンドエージェント実行中のCtrl+C/ESC無視バグの修正、起動パフォーマンスの大幅改善、シンプルモードへのファイル編集ツール追加、Yoga WASMメモリリークの修正などが含まれる。また、MaxプランにおいてSonnet 4.6が1Mコンテキストに対応した。
人気記事
ChatGPT(OpenAI)とClaude(Anthropic)の機能比較 2026年版。コーディング・長文解析・コスト・API料金の違いを検証
ChatGPT(GPT-4o/o3)とClaude(Sonnet 4.6/Opus 4.5)を2026年時点の最新情報で比較する。コーディング能力、長文処理、日本語品質、API料金、無料プランの違いをSWE-benchなどのベンチマーク結果とともに解説する。
【2026年2月20日 所感】「AIがコードを書く」は仮説から現実になった——しかし私たちはその意味をまだ消化できていない
2026年2月20日に観測したコーディングエージェント関連ニュースの総括と所感。Anthropicの自律性研究、cmux、MJ Rathbunのエージェント事故、HN「外骨格 vs チーム」論争、Stripe Minions週1000件PR、Taalas 17k tokens/sec——朝から夜までの流れを通じて見えてきた「AIがコードを書く時代」の実相を考察する。
868のスキルをnpx 1コマンドで——「Antigravity Awesome Skills」が主要AIコーディングエージェントの共通スキル基盤になりつつある
Claude Code・Gemini CLI・Codex CLI・Cursor・GitHub Copilotなど主要AIコーディングアシスタントを横断する868以上のスキルライブラリ「Antigravity Awesome Skills」(v5.4.0)を詳細分析。Anthropic・Vercel・OpenAI・Supabase・Microsoftの公式スキルを統合した設計思想、ロール別バンドル・ワークフロー機能、SKILL.mdによる相互運用性のアーキテクチャを解説する。
最新記事
AIエージェント間通信の標準化競争が始まる——AquaとAgent Semantic Protocolが同日登場
2026年2月23日、Hacker Newsに2つのAIエージェント通信プロジェクトが同日掲載された。Go製CLI「Aqua」とセマンティックルーティングを実装する「Agent Semantic Protocol」は、MCPが解決できないP2P・非同期通信の課題に取り組む。
Claude Sonnet 4.6、無料・Proプランのデフォルトモデルに——社内テストでOpus 4.5を59%の確率で上回る
Anthropicは2026年2月17日にリリースしたClaude Sonnet 4.6を、claude.aiの無料・Proプランのデフォルトモデルに設定した。価格はSonnet 4.5と同額の$3/$15 per 1Mトークン。社内評価ではコーディングエージェント用途でOpus 4.5を上回る結果が出ている。
GoogleがOpenClaw経由のGemini利用ユーザーのアカウントを永久停止——月額$250請求継続のまま
2026年2月23日、Hacker Newsで140pt/107コメントを集めたレポートによると、GoogleはOpenClaw(サードパーティクライアント)経由でGeminiを使用していたGoogle AI Pro/Ultraユーザーを予告なしに永久停止した。技術的・経済的背景を整理する。