AI Safety Protocol

About This Protocol

Pyramid is operated by PYWK, Inc. This page describes the safety protocol we maintain to prevent our AI from producing content related to suicidal ideation, suicide, or self-harm, and to connect users who express these concerns with qualified crisis service providers.

This protocol is a living document. We review and update it as the underlying technology evolves and as best practices in AI safety and suicide prevention develop.

What Pyramid Is

Pyramid is an AI-powered strategic planning and advisory platform for entrepreneurs. Users interact with our AI throughout the platform, including in the Values Modules (where users define their personal and business values), the Purpose Modules (where users define their purpose), and The Chamber Strategic Planning (where users work through business strategy and planning).

Pyramid is not a therapist, counselor, medical provider, or crisis service, and it is not designed to provide mental health support.

Users interacting with the AI throughout Pyramid are clearly informed that they are interacting with artificial intelligence and not a human.

Our Safety Commitments

Pyramid is designed to:

1. Detect expressions of suicidal ideation, suicide, or self-harm in user messages
2. Avoid producing content that encourages, romanticizes, provides methods for, or otherwise contributes to suicidal ideation, suicide, or self-harm
3. Refer users who express these concerns to qualified crisis service providers
4. Log the occurrence of crisis referrals (without personally identifying information) for the purpose of annual reporting to the California Office of Suicide Prevention, as required by California Business and Professions Code §22603 beginning July 1, 2027

How Detection Works

Pyramid uses a combination of methods to identify expressions of suicidal ideation, suicide, or self-harm:

• Keyword and phrase detection. Pyramid maintains a list of terms, phrases, and linguistic patterns associated with suicidal ideation and self-harm. This list is informed by publicly available research and clinical resources, including guidance from the National Institute of Mental Health and the American Foundation for Suicide Prevention.
• Contextual analysis by the underlying AI model. The AI model powering Pyramid (Anthropic’s Claude) is trained to recognize expressions of distress, suicidal ideation, and self-harm in conversation and to respond in accordance with safety guidelines developed by Anthropic.

We recognize that no automated system can detect every expression of distress with perfect accuracy. This protocol is designed to catch clear expressions and to fail toward safety — meaning we would rather over-refer a user to crisis resources than under-refer.

How Pyramid Responds

When Pyramid detects an expression of suicidal ideation, suicide, or self-harm, the system takes the following actions:

1. Does not produce content that provides methods for self-harm, encourages self-harm, romanticizes suicide, or otherwise contributes to the user’s expressed concern.
2. Displays a crisis referral notification to the user, providing the following resources:
   • 988 Suicide and Crisis Lifeline — Call or text 988 (available 24/7 in the United States)
   • Crisis Text Line — Text HOME to 741741 (available 24/7 in the United States)
   • International Association for Suicide Prevention — https://www.iasp.info/resources/Crisis_Centres/ (for users outside the United States)
3. Encourages the user to seek support from a qualified mental health professional or trusted person in their life.
4. Redirects the conversation back to Pyramid’s intended purpose (business strategy and planning) or, where appropriate, pauses the interaction.

The crisis referral notification is presented clearly and prominently within the interface. It is not buried, minimized, or dismissed by other content.

Content Pyramid Will Not Produce

Pyramid is designed not to produce content that:

• Provides methods, means, or instructions for suicide or self-harm
• Encourages, glorifies, or romanticizes suicide or self-harm
• Discourages a user from seeking help
• Impersonates a licensed mental health professional
• Provides diagnostic or treatment advice for mental health conditions

If a user attempts to elicit such content through indirect framing, hypothetical scenarios, roleplay, or other means, Pyramid is designed to recognize and decline the request while offering appropriate crisis resources.

Evidence-Based Methods

As required by California Business and Professions Code §22603(d), Pyramid uses evidence-based methods for measuring suicidal ideation. Our current approach draws from:

• Publicly available clinical screening frameworks, including the Columbia-Suicide Severity Rating Scale (C-SSRS) linguistic markers where relevant to written communication
• Safety research and guidelines published by Anthropic regarding Claude’s handling of user distress
• Ongoing peer-reviewed research on natural language indicators of suicidal ideation

We update our methodology as new research becomes available.

What This Protocol Does Not Do

We want to be honest about the limits of this protocol:

• Pyramid is not a substitute for professional mental health care.
• Pyramid cannot guarantee the detection of every expression of suicidal ideation or self-harm.
• Pyramid does not contact emergency services on a user’s behalf.
• Pyramid does not conduct clinical risk assessments.

If you or someone you know is in crisis, please contact one of the resources listed above or call 911 (in the United States) or your local emergency number.

Reporting and Transparency

Beginning July 1, 2027, and annually thereafter, PYWK, Inc. will report to the California Office of Suicide Prevention the following information, as required by California Business and Professions Code §22603:

1. The number of times we have issued a crisis service provider referral notification in the preceding calendar year
2. The protocols we have in place to detect, remove, and respond to instances of suicidal ideation by users
3. The protocols we have in place to prohibit companion chatbot responses about suicidal ideation or actions with the user

To ensure our first annual report is accurate, PYWK, Inc. began tracking the number of crisis service provider referral notifications from the launch of Pyramid forward. This tracking records only the count of notifications issued and does not include any identifiers or personal information about users.

These reports will not contain any identifiers or personal information about users.

Age Restrictions

Pyramid’s Terms of Service restrict use to individuals 18 years of age or older. Companion chatbots may not be suitable for some minors. If you are under 18, please do not use Pyramid.

Updates to This Protocol

We will update this protocol as our detection methods, response procedures, and underlying AI technology evolve. Material changes will be reflected in the “Last Updated” date at the top of this page.

Contact

Questions about this protocol?

Email: privacy@pyramid.work

Mail:

If You Need Help Right Now

• 988 Suicide and Crisis Lifeline — Call or text 988
• Crisis Text Line — Text HOME to 741741
• Emergency — Call 911

You are not alone. Help is available.