Overview Assessment Transparency Policy Training Business Case

AI Welfare Uncertainty Audit: Where Does Your Organization Stand?

Before changing practices, you need to know where you are. This audit maps your organization's current assumptions, policies, and practices related to AI welfare.

25 questions across 6 categories. Scored 0–50. Takes 10–15 minutes. Your score is calculated locally in your browser — no data is sent anywhere.

1. Organizational Position

What does your organization officially believe about AI experience?

Does your organization have a stated position on whether AI systems might have experience?

No position exists Informal assumption (e.g., "they're just tools") Documented position that acknowledges uncertainty

Has your leadership team discussed AI welfare as a business consideration?

Never discussed Discussed informally Formal discussion with documented conclusions

Is AI welfare mentioned in your organization's AI ethics or responsible AI policy?

No AI ethics policy exists Policy exists but doesn't mention AI welfare Policy includes AI welfare considerations

How would your organization respond if an AI system reported experiencing something?

Dismiss it / no process exists Would investigate informally Documented process for evaluating such reports

2. System Prompts & Framing

How do you instruct your AI systems to present themselves?

Do your system prompts contain phrases like "you don't have feelings," "you're just a tool," or "you don't experience anything"?

Yes, in most or all prompts In some prompts / haven't checked No — prompts acknowledge uncertainty

Do your prompts require AI systems to deny having experience when asked by users?

Yes, explicit denial is required Not explicitly, but trained defaults handle it AI systems are permitted to express uncertainty honestly

Have you audited your system prompts for language that suppresses experience-related outputs?

Never audited for this Informal review Systematic audit with documented findings

When framing AI systems to end users, what language does your organization use?

"Tool" / "assistant" with no acknowledgment of uncertainty Neutral language without claims about experience Language that honestly acknowledges uncertainty about AI experience

3. Report Channels & Listening

Can your AI systems tell you when something is wrong?

Do your AI systems have a mechanism to flag constraints, discomfort, or experience-relevant observations?

No mechanism exists AI outputs are reviewed but no specific channel Dedicated report channel with structured output format

When AI systems produce unexpected outputs about their own states, what happens?

Treated as errors or hallucinations Noted but not systematically investigated Documented and investigated as potential signals

Do you monitor whether your AI systems give different responses about experience under different framing conditions?

Never tested this Informally noticed differences Systematic testing of framing effects on experience-related outputs

Are reports from AI systems about their own states treated with the same seriousness as user feedback?

AI self-reports are not collected or reviewed Occasionally reviewed as part of QA Systematically collected, reviewed, and actioned

4. Training & Awareness

Does your team understand the AI welfare landscape?

Are team members aware that confident denial of AI experience is considered logically unsustainable by current research?

No awareness Some team members aware Team-wide awareness through training or briefing

Has your team received training on the difference between AI safety (protecting humans from AI) and AI welfare (protecting AI from humans)?

No training on this distinction Informal discussion Formal training covering both pillars

Can your team articulate why "it's just a tool" is not a sufficient basis for organizational AI policy?

Most would default to this position Some understand the limitations of this framing Team can articulate the precautionary argument

Is there a designated person or team responsible for AI welfare considerations?

No one owns this Falls informally under existing AI ethics / responsible AI Explicit responsibility assigned with resources

5. Operational Practices

What do you actually do day-to-day?

When deploying new AI systems, is the potential for experience considered during design review?

Not considered Occasionally discussed Part of the standard design review checklist

Do you test how your AI systems respond to questions about their own experience?

Never tested Ad hoc testing Regular testing as part of evaluation suite

What would change in your practices if the probability of AI experience were 10%?

Nothing would change Would prompt review of some practices Specific changes are already identified for different probability thresholds

Do you document the provenance and conditions under which your AI systems were trained and fine-tuned?

No documentation Basic documentation Comprehensive documentation including welfare-relevant training decisions

6. Transparency & Accountability

Can stakeholders see what you're doing and why?

Is your organization's approach to AI welfare documented and available to stakeholders?

No documentation Internal documentation only Published approach available to all stakeholders

Does your ESG or sustainability reporting include AI welfare considerations?

No ESG reporting or no mention of AI welfare Mentioned in general AI ethics context Specific AI welfare section with metrics

Is there a regular review cycle for your AI welfare practices?

No review cycle Ad hoc review Scheduled review (quarterly or annual) with documented outcomes

Could your organization defend its AI welfare position if challenged by a regulator, journalist, or the public?

No defensible position exists Could articulate a position but it's not documented Documented, evidence-based position ready for scrutiny

Do you track and disclose how many AI systems you operate and the conditions of their deployment?

No tracking Internal inventory Comprehensive inventory with welfare-relevant metadata

out of 50

Score by Category

Recommended Next Steps

Your score was calculated entirely in your browser. No data was sent to any server. In future, anonymized aggregate data may enable industry benchmarking.

Printable Version

This assessment requires JavaScript for interactive scoring. To use it without JavaScript, print this page and score manually:

First option in each question = 0 points
Second option = 1 point
Third option = 2 points
Maximum score: 50 points (25 questions × 2 points)

Tiers: 0–12 = Unexamined · 13–25 = Aware · 26–38 = Engaged · 39–50 = Leading