OpenClaw Provider Policies: Which APIs Allow It

Evidence-based analysis of provider terms and enforcement patterns.

TL;DR

Methodology

This verification analyzes:

1. Primary source terms — Official ToS, acceptable use policies, legal pages
2. Enforcement history — Documented actions against users
3. Product patterns — What the provider builds reveals intent
4. Community signals — Forum reports, GitHub issues, social media

Evidence levels:

• High: Official policy documents, direct quotes, dated
• Medium: Help center guidance, standard terms analysis, product behavior
• Low: Community reports, indirect signals

Risk Spectrum

BAN RISK SPECTRUM — OpenClaw API Usage
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔴 BAN LIKELY              🟡 MEDIUM RISK            🟢 SAFE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Anthropic              Groq*                   Google/Kimi/Together/
(Explicit ban)         (Orchestration          Fireworks/OpenAI/Bedrock/
                        clause caveat)          Self-Hosted
Google OAuth           Azure OpenAI*
(OAuth blocked)        (Governance required)

* Medium risk = allowed but with specific caveats to follow

Key insight: Only Anthropic and Google OAuth have explicitly banned OpenClaw-like usage with documented enforcement. All others either allow it or haven’t addressed it. See the full migration guide if you’re using a banned provider.

Provider Analysis

Anthropic: Explicitly Disallowed 🔴

Status: 🚫 DISALLOWED — High Ban Risk

The Policy (Primary Source):

“Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.” — Anthropic Claude Code Legal and Compliance, February 2026

Enforcement History:

Detection Methods:

• Client fingerprinting via HTTP headers
• Pattern analysis of request volume/timing
• Binary signature detection

Evidence Level: HIGH (official policy + technical enforcement)

Deep Dive: /verify/openclaw-anthropic-policy/

Google: Split Status

Google OAuth (Antigravity): Disallowed 🔴

Status: 🚫 DISALLOWED — High Ban Risk

The Policy:

“You must not abuse, harm, interfere with, or disrupt the Service. This includes, but is not limited to, using the Service in connection with products not provided by us.” — Antigravity Additional Terms of Service, Section 6

Enforcement History:

Scope: Only Antigravity access suspended (Gmail/Workspace unaffected per Google)

Recovery (March 2026): Recovery path confirmed for unaware first-time violators. Contact [email protected] — 1-2 week response time reported.

Evidence Level: HIGH (official statements + mass enforcement + confirmed recoveries)

Deep Dive: /verify/openclaw-google-policy/

Google Gemini API: Allowed with Caution 🟡

Status: ✅ ALLOWED — Low-Medium Risk

Analysis:

• ✅ No explicit ban on third-party agent tools in API terms
• ✅ Google AI Studio / Vertex AI — standard acceptable use policies
• ⚠️ Google’s platform control philosophy — OAuth banned, API allowed… for now
• ⚠️ Monitor for policy changes — Google has shown willingness to restrict

Evidence Level: MEDIUM (absence of prohibition, but provider pattern suggests caution)

Data Loss Risk:

• API keys are commercial contracts (more stable than consumer OAuth)
• But Google’s enforcement history shows sudden policy shifts
• Recommendation: Maintain local backups, don’t rely solely on Gemini API

OpenAI: Explicitly Allowed 🟢

Status: ✅ ALLOWED — Low Risk

The Evidence:

OpenAI’s own Codex CLI explicitly allows signing in with ChatGPT Plus/Pro subscriptions via OAuth. This reveals architectural intent: subscription credits can flow to third-party CLI tools.

Why this matters:

• Codex CLI is an official OpenAI product
• It uses the exact pattern Anthropic banned (OAuth → CLI automation)
• No enforcement against agent tools documented

The Only Caveat:

OpenAI’s terms reserve the right to suspend for “unusual activity patterns.” However:

1. No agent tool bans reported
2. Codex enables automation by design
3. Usage matches their product design

Evidence Level: HIGH (product behavior proves policy intent)

Deep Dive: /verify/openclaw-openai-policy/

Moonshot AI (Kimi k2.5): Allowed 🟢

Status: ✅ ALLOWED — Low Risk

Why Kimi is low-risk:

• ✅ Open-source weights — Can self-host if API terms change
• ✅ No restrictions — Moonshot AI promotes agentic use cases
• ✅ API terms standard — No tool-specific prohibitions

Evidence Level: HIGH (open-source + explicit support for agents)

AWS Bedrock: Allowed 🟢

Status: ✅ ALLOWED — Low Risk

Analysis:

• Standard AWS acceptable use policy
• No tool-specific restrictions
• Enterprise-focused, automation expected

Evidence Level: MEDIUM (standard AWS terms + enterprise orientation)

Together AI: Allowed 🟢

Status: ✅ ALLOWED — Low Risk

Analysis:

• Inference provider for open-source models
• BYOK (bring your own key) model
• No restrictions on client tools found

Evidence Level: MEDIUM (inference provider model)

Fireworks AI: Allowed 🟢

Status: ✅ ALLOWED — Low Risk

Analysis:

• Inference provider, similar to Together
• Open-source model hosting
• No tool restrictions in terms

Evidence Level: MEDIUM (inference provider model)

Groq: Allowed with Caveats 🟡

Status: ✅ ALLOWED — Medium Risk

Analysis:

• ✅ No explicit ban on third-party agent tools
• ⚠️ “Orchestration” clause prohibits coordinating usage across multiple organizations
• ⚠️ High-speed inference can trigger “excessive load” concerns
• Rate limits are actively enforced

Key Policy Quote:

“orchestrating usage between multiple organizations”

Evidence Level: MEDIUM (no ban + vague clause that could catch automation)

Safe Usage:

• Use single API key per OpenClaw instance
• Stay within published rate limits
• Don’t coordinate across multiple Groq accounts

Deep Dive: /verify/openclaw-groq-policy/

Azure OpenAI: Allowed with Governance 🟡

Status: ✅ ALLOWED — Medium Risk (Governance-dependent)

Analysis:

• ✅ No ban on autonomous applications — Code of Conduct explicitly mentions them
• ⚠️ Strict governance requirements — Must implement “technical and operational measures”
• ✅ Enterprise Agreement advantage — Contract terms govern for EA customers
• ✅ Limited Access service — Most customers eligible for standard access

Key Policy Quote:

“applications that make decisions, or take actions, autonomously or with varying levels of human intervention”

Evidence Level: MEDIUM (explicit autonomous app language + strict compliance requirements)

Requirements:

• Implement access controls (Azure Key Vault for API keys)
• Enable audit logging (Azure Monitor)
• Provide human oversight for consequential decisions
• Maintain “strong technical controls”

Deep Dive: /verify/openclaw-azure-policy/

Cerebras: Allowed 🟢

Status: ✅ ALLOWED — Low Risk

Analysis:

• ✅ Infrastructure provider model — No client-side restrictions
• ✅ No explicit ban on third-party agent tools
• ✅ Focus on credential security, not usage patterns
• Similar to Together AI and Fireworks AI

Evidence Level: MEDIUM (standard infrastructure terms)

Deep Dive: /verify/openclaw-cerebras-policy/

Perplexity: Allowed with Caveats 🟡

Status: ✅ ALLOWED — Low-Medium Risk

Analysis:

• ✅ No explicit ban on third-party agent tools
• ⚠️ Competitive use clause — Don’t build search products
• ⚠️ “Interfering with” clause — Respect rate limits
• Safe for coding workflows, risky for search aggregation

Evidence Level: MEDIUM (API terms + AUP compliance required)

Deep Dive: /verify/openclaw-perplexity-policy/

Mistral AI: Explicitly Supported 🟢

Status: ✅ ALLOWED — Low Risk

Analysis:

• ✅ Explicit third-party support — Terms permit Third-Party Services
• ✅ MCP servers named — Protocol OpenClaw uses is specifically allowed
• ✅ Open-source models — Can self-host if needed
• European provider with interoperability-friendly stance

Evidence Level: HIGH (explicit permission + MCP naming)

Deep Dive: /verify/openclaw-mistral-policy/

Self-Hosted: Always Allowed 🟢

Status: ✅ ALLOWED — No Risk

The Ultimate Escape Hatch:

• Run models locally via Ollama, llama.cpp, vLLM
• No API provider = no terms to violate
• You are the provider

Trade-offs:

Evidence Level: N/A (no third-party terms)

Provider Philosophy Comparison

The Lesson: Precedent is policy. Look at what vendors do, not just what they say.

What Would Change These Ratings

To upgrade a provider (safer):

• Official statement authorizing third-party tool usage
• Published policy explicitly allowing OpenClaw-like patterns
• No enforcement actions for 12+ months

To downgrade a provider (riskier):

• Policy update restricting third-party tools
• Documented enforcement action
• Technical blocks deployed

Review cadence: Monthly during policy volatility; quarterly otherwise.

FAQ: Can I Use OpenClaw With My Provider?

Quick Answers

What Could I Lose If I Get Banned?

With Anthropic (Claude):

• ❌ All conversation history — Gone permanently, no export during suspension
• ❌ Claude Memories — Personalized context deleted
• ❌ Projects/Artifacts — Any unsaved work inaccessible
• ❌ API access — Keys revoked, commercial contracts voided
• ⚠️ Account recovery — Possible but timeline uncertain (1-2+ weeks)

With Google (Antigravity OAuth):

• ❌ Antigravity IDE access — Suspended immediately (403 ToS error)
• ❌ Chat history — Inaccessible during suspension, no self-service export
• ✅ Gmail/Workspace — Unaffected (Google confirms surgical targeting)
• ⚠️ Billing — May continue during suspension
• ⚠️ Recovery path — Exists for “unaware users” but timeline unknown

With OpenAI, Kimi, Mistral (Safe Providers):

• ✅ Nothing — These providers explicitly allow or don’t restrict third-party tools
• ✅ No enforcement history — Zero documented bans for agent tool usage
• ✅ Rest easy — Your setup is compliant

I’m Currently Using [Provider X]. What Should I Do?

If Using OpenAI

Status: ✅ You’re safe. Rest easy.

Recommended actions:

1. No changes required
2. Keep using ChatGPT OAuth via Codex CLI or API keys
3. Monitor OpenAI policy updates monthly
4. Maintain backup provider (Kimi/Google) for redundancy

If Using Kimi

Status: ✅ You’re safe. Rest easy.

Recommended actions:

1. No changes required
2. Enjoy 8x cheaper pricing than Anthropic
3. Consider self-hosting option (weights are open-source)

If Using Google (Gemini API Keys)

Status: ✅ You’re safe, but monitor policy.

Recommended actions:

1. Verify you’re using API keys (from Google AI Studio), NOT Antigravity OAuth
2. Maintain local backups of critical conversations
3. Keep alternative provider ready (OpenAI/Kimi)
4. Set up billing alerts — charges can accumulate quickly

If Using Google (Antigravity OAuth)

Status: 🚫 STOP IMMEDIATELY — You’re violating ToS.

Immediate actions:

1. STOP using Antigravity OAuth with OpenClaw — Every use increases ban risk
2. Export your data — While you still have access (Settings → Data Export)
3. Switch to Gemini API keys — Generate at Google AI Studio
4. Revoke OAuth authorization — Google Account → Security → Third-party apps
5. Test new setup — Verify OpenClaw works with API keys before relying on it

If Using Anthropic

Status: 🚫 STOP IMMEDIATELY — You’re violating ToS.

Immediate actions:

1. STOP using Anthropic OAuth with OpenClaw — Ban risk is HIGH
2. Export your data NOW — Claude.ai → Settings → Data Export (takes 24-48 hrs)
3. Screenshot critical conversations — No bulk export for individual chats
4. Migrate to safe provider — See migration paths below

Migration Paths (If You Need to Switch)

Recommended: OpenAI

• Why: Explicitly allows OAuth via Codex CLI, no enforcement history
• Cost: ChatGPT Plus ($20/mo) or API pay-as-you-go
• Migration time: 15 minutes
• Guide: /verify/openclaw-openai-policy/

Recommended: Kimi k2.5

• Why: 8x cheaper than Anthropic ($3 vs $25 per 1M output tokens), no restrictions
• Cost: API or free tier via OpenCode Zen
• Migration time: 10 minutes
• Guide: /tools/opencode/

Budget Option: Self-Hosted

• Why: Zero provider risk, you are the provider
• Cost: Upfront hardware cost only
• Migration time: 1-2 hours setup
• Guide: /tools/self-hosting/

How Do I Know If I’m at Risk?

You’re at HIGH risk if:

• You authenticated OpenClaw with “Sign in with Google” through Antigravity
• You authenticated OpenClaw with Claude (Free/Pro/Max) OAuth
• You see 403 errors or “Terms of Service” messages
• You received a warning email from Anthropic or Google

You’re at LOW risk if:

• You use OpenAI with ChatGPT OAuth or API keys
• You use Kimi with API keys
• You use Mistral with API keys
• You use Google Gemini API keys (NOT Antigravity)

You’re at MEDIUM risk if:

• You use Google Gemini API keys (provider has shown policy volatility)
• You use Groq (orchestration clause caveat)
• You use Azure OpenAI (governance requirements)

What’s the Safest Setup?

Maximum Safety (Zero Provider Risk):

Self-hosted models (Ollama/llama.cpp)
├─ No terms to violate
├─ No API provider to ban you
└─ Trade-off: Setup complexity, limited to open models

High Safety (Explicitly Allowed):

OpenAI (ChatGPT OAuth or API keys)
├─ Codex CLI proves third-party OAuth is intended
├─ No enforcement history
└─ Best frontier model access

High Safety (Cost-Effective):

Kimi k2.5 (API keys)
├─ Open-source weights (can self-host)
├─ 8x cheaper than Anthropic
└─ No restrictions found

Related Links

News & Context:

• /posts/openclaw-provider-policy-check-2026/ — Breaking news, migration paths, enforcement timeline
• /posts/anthropic-claude-code-oauth-policy-feb-2026/ — Anthropic enforcement analysis

Risk Assessment:

• /risks/google/antigravity-oauth-risk/ — Google-specific risk
• /risks/anthropic/third-party-access/ — Anthropic risk factors
• /risks/anthropic/account-bans/ — Account ban patterns

Terms Verification:

• /verify/openclaw-anthropic-policy/ — Anthropic OpenClaw policy (BANNED)
• /verify/openclaw-openai-policy/ — OpenAI OpenClaw policy (ALLOWED)
• /verify/openclaw-groq-policy/ — Groq OpenClaw policy (ALLOWED with caveats)
• /verify/openclaw-azure-policy/ — Azure OpenAI policy (ALLOWED with governance)
• /verify/openclaw-cerebras-policy/ — Cerebras OpenClaw policy (ALLOWED)
• /verify/openclaw-perplexity-policy/ — Perplexity OpenClaw policy (ALLOWED with caveats)
• /verify/openclaw-mistral-policy/ — Mistral AI OpenClaw policy (ALLOWED)
• /verify/openclaw-google-policy/ — Google policy (OAuth banned, API allowed)
• /verify/claude-code-terms/ — Anthropic terms verification

Migration & Alternatives:

• /tools/opencode/ — Free Kimi k2.5 access
• /value/free-stack/ — Complete free tier guide
• /tools/self-hosting/ — Zero-provider-risk option

Sources

Primary Sources

• Anthropic Claude Code Legal and Compliance — code.claude.com/docs/en/legal-and-compliance — Official policy, February 2026 (Archive lookup)
• Varun Mohan (Google DeepMind) — X post — Enforcement confirmation, February 2026 (Archive lookup)
• Google AI Developer Forum — Mass 403 Bans thread — User reports, February 2026 (Archive lookup)
• OpenClaw GitHub — Issue #14203 — Ban wave documentation (Archive lookup)

Terms of Service

• Anthropic Consumer Terms (Archive lookup)
• Google Gemini API Terms (Archive lookup)
• OpenAI Terms of Service (Archive lookup)
• Moonshot AI Platform Terms (Archive lookup)

Last updated: 2026-02-25
Next review: 2026-03-24
Evidence level: High (primary sources + dated)

Policy pages change. Verify current terms at official sources before making architectural decisions.