Spec27 bot
Spec27 bot researches websites and tests AI agents. Safe Intelligence Ltd. operates the bot for Spec27 users and our research team.
What Spec27 bot does
Website research
Spec27 retrieves product information, help articles, policies and service descriptions. We use this content to prepare evaluation datasets and check AI-agent responses against published information.
AI-agent testing
Spec27 runs automated conversations through AI agents' web interfaces and APIs. Tests include ordinary questions, follow-up messages and adversarial prompts that help identify incorrect answers and unsafe behaviour.
Tests can create chat sessions and submit messages that appear in your chat analytics and conversation history.
Identify Spec27 requests
Our HTTP User-Agent contains Spec27/1.0:
Mozilla/5.0 (X11; Linux x86_64; Spec27/1.0) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/142.0.0.0 Safari/537.36
The browser version may change. Match Spec27/1.0 to recognize this bot version.
Identify a specific execution
Outbound sandbox requests include a Referer header identifying the execution
that sent them. For the Spec27-hosted service, it has this format:
https://dashboard.spec27.ai/orgs/{organization_id}/sandbox/executions/{execution_id}
Include the complete Referer value in abuse reports so we can trace the request
to its execution.
Verify a request
Spec27 signs requests with Ed25519 HTTP message signatures using
Web Bot Auth.
Signed requests include Signature, Signature-Input and Signature-Agent
headers.
The Spec27-hosted service uses this signing identity:
https://dashboard.spec27.ai
Verify signatures against the public keys at:
https://dashboard.spec27.ai/.well-known/http-message-signatures-directory
Control access
Use Spec27 in your robots.txt rules. To exclude your entire website:
User-agent: Spec27
Disallow: /
To exclude a particular path:
User-agent: Spec27
Disallow: /private/
Publish the rules in /robots.txt on each origin you want to protect.
Spec27 checks robots.txt before making HTTP requests, including redirects. Rules are cached for up to 24 hours, or less when your HTTP caching headers require it. Changes take effect when the cached rules are refreshed.
If robots.txt is temporarily unavailable, previously retrieved rules may remain in use. Use your server or firewall access controls to block requests immediately. Contact us to request removal of information already collected.
How we use collected information
We use website content and agent responses to prepare evaluation datasets and assess agent behaviour. AI models may process this material during evaluations.
We do not use content collected by Spec27 bot to train or fine-tune general-purpose AI models.
Our research may include published company profiles, evaluation examples and benchmark results. Contact support@spec27.ai with questions about data use, retention or removal.
Report unwanted activity or abuse
Email support@spec27.ai with the subject "Spec27 bot report" and include:
- The complete
Refererheader, if present, to identify the execution. - The affected domain and URL.
- The date, time and time zone.
- The observed User-Agent and source IP address.
- A request ID or Cloudflare Ray ID, if available.
- The activity you want us to investigate.
Remove passwords, access tokens, cookies and personal conversation content from logs before sending them.
Operator
Contact: support@spec27.ai