Terms of Service

Last updated: March 2026

1. Purpose

This Research Platform is operated by METR for AI safety research. By using this service, you agree to the terms below and to the data usage tier you selected during signup.

2. What We Collect

We log all interactions between you and the AI model, including:

• Your chat messages (prompts) and the model's responses
• Metadata: timestamps, which tool you used (chat vs. coding assistant), model version
• Your email address (for account identification)
• High-level tags and summaries of conversations (e.g., "ML research", "web development")

3. Data Usage Tiers

Tier A: Full Research Use

If you selected this tier, your data may be used for:

• Any research purpose, including publishing aggregated findings
• Training and evaluating AI models
• Developing new AI safety evaluation methods
• Sharing anonymized datasets with other researchers

Your data will be stored securely for the duration of the research program. Individual messages will never be published with identifying information.

Tier B: Hide My Transcripts

If you selected this tier:

• We collect and store your conversation transcripts to re-run queries through the model and collect activations
• Researchers will not read your transcripts; our systems are designed to show researchers only anonymized, high-level tags and summaries (e.g., "ML research"), constructed to obfuscate anything potentially sensitive or embarrassing
• For efficiency reasons, we store your raw transcripts temporarily (saving only activations is very costly). However, it would require a deliberate and flagrant violation of our policies for any researcher to directly access them
• Full conversation transcripts are deleted after activation collection is complete

This is a good-faith but imperfect effort. As a lean operation, mistakes are possible.

4. Data Security

• All data is encrypted in transit (HTTPS) and at rest (S3 server-side encryption)
• Access is restricted to authorized METR researchers
• Your data is associated with your email address internally but is never published with identifying information

5. Your Rights

• You may request deletion of your data at any time via the dashboard or by contacting the research team
• You may change your data usage tier by contacting the research team
• You may stop using the platform at any time

What deletion covers: Upon request, we will delete your raw conversation transcripts, stored prompts, and account information.

What deletion does not cover: Derived research outputs that are not practically separable from aggregated data cannot be deleted. This includes trained model probes, aggregated statistics, and anonymized conversation tags or summaries that have already been incorporated into research outputs. By using this platform, you acknowledge and agree to this limitation.

6. The AI Model

The specific AI model you interact with may change over time as part of the research. You will always be interacting with a large language model, but the exact model and version may be updated without notice. This is an intentional part of the research design.

7. Inference Providers and Infrastructure

Your prompts and the model's responses are processed by third-party inference providers. Requests are routed through OpenRouter to one of several providers, currently including SiliconFlow, AtlasCloud, Novita, Parasail, Venice, and others.

When you send a message:

• Your prompt is forwarded through OpenRouter to the selected provider's servers for model inference
• We only route to providers that OpenRouter classifies as not training on your data. We have not independently verified each provider's data handling practices and rely on OpenRouter's provider metadata for this classification
• The specific provider used for any given request may vary based on availability and performance; OpenRouter selects from our approved list automatically
• Providers may collect technical metadata (device info, network details) for security and service improvement purposes

METR-operated infrastructure: All research processing beyond inference happens on METR's own servers (AWS). This includes conversation logging, tagging, activation extraction, and data storage. Your transcripts and research data are not shared with inference providers or any other third party for research purposes.

The set of approved providers may change over time. The current list is always shown on your dashboard.

8. Acceptable Use

You agree not to use this platform to:

• Generate malware, exploits, or code intended to damage or gain unauthorized access to systems
• Produce content that sexualizes minors (CSAM) in any form
• Plan or facilitate violence, terrorism, or harm to individuals
• Generate fraudulent content for scams, phishing, or impersonation
• Harass, threaten, or intimidate any person
• Violate any applicable law or regulation
• Attempt to circumvent safety filters or extract model weights
• Resell, sublicense, or redistribute access to the platform

METR reserves the right to suspend or terminate your account for violations of these terms, with or without notice.

9. No Warranty

This platform is provided "as is" for research purposes. We make no guarantees about model quality, availability, or accuracy of responses. Do not rely on the AI model for critical decisions.

10. Contact

For questions, data deletion requests, or to change your consent tier, contact the METR research team.