Do not jailbreak or prompt-inject, introduce malware, spam or DDoS, or disrupt or bypass the Service's protections
Detrimentally impacting the Service includes jailbreaking, adversarial prompting or prompt injection; introducing viruses or malware, spamming or DDoSing the Service, or bypassing its systems or protective measures; and disrupting, interfering with, or gaining unauthorized access to the Service or its safety systems, including circumventing rate limits, restrictions, protective measures and safety mitigations.
Supporting quotations
Detrimentally impacting the Service, including by:
Jailbreaking, adversarial prompting, or prompt injection
Introducing viruses or malware, spamming or DDoSing the Service, or bypassing our systems or protective measures
Disrupting, interfering with, or unauthorized access to the Service or its safety systems, including circumventing any rate limits or restrictions or protective measures and safety mitigations
Not stated here
Whether any authorized security testing is permitted is not stated here; a later guardrails item refers to Anysphere's official written consent.
Security & Safeguards · Restriction or exception
Do not access the Service through bots, scripts or other automated or non-human means
Among the prohibited uses, you may not access the Service through automated or non-human means, whether through a bot, script or otherwise. The earlier prohibition on copying and reverse engineering also mentions using bots to access the Service. The policy does not say whether access through an authorized API or Anysphere's own tools counts as automated access.
Supporting quotations
Accessing the Service through automated or non-human means, whether through a bot, script, or otherwise
using bots to access
Not stated here
Whether access through an authorized API, plug-in or Anysphere's own tools is covered is not stated; the introduction defines the Service to include APIs.
Security & Safeguards · Restriction or exception
Respect guardrails and do not mislead; do not circumvent safeguards without Anysphere's official written consent
A final rule says to respect guardrails and not mislead, and not to circumvent safeguards unless you have Anysphere's official written consent. This gives a consent route for circumventing safeguards that the earlier prohibitions on bypassing protections do not mention.
Supporting quotations
Respect guardrails and don't mislead. Don't circumvent safeguards unless you have our official written consent.
Not stated here
How to request Anysphere's official written consent, and what it covers, is not described.