More than our method: a governance layer any AI system can adopt. If an AI touches people's safety, trust, or information, we believe it owes them these six commitments.
P · Protection
The system's first duty is to the person in front of it, not to engagement, data collection, or upsell. Protective action is proactive: warn early, before harm, in language the person can act on.
R · Respect
Privacy by default. No advertising use of protective data, no sale of personal information, and transient processing wherever possible. The person chooses what leaves their device, and consent is explained in plain words, not legal fog.
I · Integrity
Every verdict shows its evidence. No black-box judgments: a protective AI must be able to say what it checked, what it found, and how it decided, in human-readable form. If it cannot support a claim, it says so instead of guessing.
N · Neutrality
The system detects manipulation patterns and coordinated deception, never ideology. It does not decide what people should believe; it shows how something is trying to influence them and lets the human decide.
C · Clarity
Safety must survive translation. Alerts, evidence, and next steps are available in the person's own language and their own channel: read, heard, or felt. Accessibility is a floor, not a feature.
E · Empowerment
The human makes the final call. The system educates, explains, and offers recovery paths; it does not seize control. Every alert should leave the person more capable than before it fired.
Four lines we hold ourselves to
Added in September 2026. These say how the six commitments meet a real person.
A guardian's success is measured by what the person can do without it. A guardian gets you to a real person, a next step, or your own judgment, and then it leaves. It never manufactures a reason to stay.
Capability and authority are separate variables, and accountability decides where authority belongs. A smarter model earns a bigger job, not more trust. No model gets unconditional trust, including our own: evidence establishes claims, permissions bound actions, and people keep authority, because people carry the consequences.
Warmth carries the evidence and never replaces it, and warmth is never a reason to stay. A guardian may be kind enough that a frightened kid or a confused grandmother asks for help. Trust it because it can show you why, not because you like it.
A guardian observes, investigates, recommends, and asks before it acts. It acts on its own only under the person's prior consent, deterministic rules, and evidence that has cleared the required threshold. No model may rewrite the rules that constrain it.
Project PRINCE is the first implementation of this protocol. It will not be the last: the charter is open for any system, company, or agency to adopt, and we welcome the conversation.
