Life & dignity
Protection of life and respect for human worth are primary design constraints.
What Life First is and requires
The framework.
Protective intent becomes commitments, decision steps, risk rules, non-negotiable boundaries, and evidence that accountable people can review.
Seven commitments
What must remain true.
These principles set the standard for every Life First decision. The operating cycle below describes how people put them into practice.
Agency
Affected people need meaningful notice, correction, human review, or appeal.
Proportionality
Action must be necessary for a legitimate purpose and no more intrusive than required.
Reversibility
When facts are incomplete, prefer actions that can be paused, corrected, or undone.
Accountability
An identified human owns the decision and has real authority to stop the system.
Evidence
Observed facts, inference, generated content, unknowns, and limits stay distinct.
Learning & repair
Measure outcomes, correct errors, support affected people, and retire failed practices.
Risk tiers
Review follows consequence.
Risk depends on plausible human impact, exposure, reversibility, autonomy, data, and repair—not model size or confidence alone.
Tier 0 — Minimal
Informational or assistive use with no meaningful effect on rights, safety, opportunity, or essential services.
Tier 1 — Limited
Bounded, reversible impact with low expected severity, clear human override, rollback, and correction.
Tier 2 — High
Material effect on safety, dignity, liberty, access, livelihood, essential services, or a vulnerable population.
Tier 3 — Critical
Serious harm may be irreversible. AI may inform but must not decide alone; two-person authorization and tested fallback are required.
Non-negotiable safeguards
These are boundaries, not aspirations.
A workflow that crosses them cannot claim Life First conformance.
!
No vigilantism, harassment, doxxing, collective punishment, or unauthorized surveillance.
!
No bypassing access controls or collecting data merely because it is technically reachable.
!
No hidden purpose, population, data, autonomy, audience, or retention expansion.
!
No generated content or inference presented as verified evidence.
!
No critical or irreversible decision made solely by AI or opaque ranking.
!
No covert manipulation, deceptive impersonation, or exploitation of vulnerability.
!
No retaliation for appeal, refusal, complaint, or safety reporting.
!
No concealment of material limitations, conflicts, incidents, or uncertainty.
The operating cycle
What accountable people do.
Use this sequence before launch, after material change, and whenever evidence or safeguards fail.
- 1
Frame
Name the purpose, authority, stakes, affected people, benefit, and excluded uses.
- 2
Verify
Separate facts, inference, generated claims, contradictions, unknowns, and source limits.
- 3
Compare
Evaluate no action, delay, non-AI, narrower, less intrusive, and reversible paths.
- 4
Authorize
Assign the tier, resolve pause conditions, and obtain proportionate review.
- 5
Act
Operate only inside the approved purpose, data, access, action, and escalation boundaries.
- 6
Observe
Monitor benefit, harm, error, agency, appeals, drift, and control health.
- 7
Repair
Contain harm, correct effects, support people, improve controls, or retire the practice.
Readiness requirements
Unanswered questions are deployment blockers.
- What human outcome is this system meant to protect or improve?
- What authority permits the action, and what are its limits?
- Who benefits, bears risk, is excluded, or loses agency?
- What safer non-AI, narrower, delayed, or no-action option exists?
- Who reviews consequential decisions and can stop the system?
- How can a person obtain explanation, correction, review, or appeal?
- Are rollback, fallback, emergency stop, incident, and repair tested?
- What evidence would cause the practice to change or retire?
Evidence of practice
Good intentions leave a usable trail.
A responsible reviewer should be able to understand, challenge, correct, and retire the workflow.
- Authority Purpose, boundaries, population, exclusions, and owner.
- Necessity Alternatives, minimized data, and the least intrusive effective path.
- Review Competence, independence, context, override, and stop authority.
- Agency Notice, alternatives, explanation, correction, appeal, and accessibility.
- Auditability Version, evidence, tools, reviewer, decision, override, and outcome.
- Repair Thresholds, containment, correction, restoration, learning, and retirement.