{"id":"d4edbfb2-32b2-48d5-a0e1-5b484a604a96","shortId":"8ASkaE","kind":"skill","title":"ai-ethics-tradeoffs","tagline":"Framework for navigating AI safety, ethics, and capability trade-off discussions. Covers responsible scaling, content policy, bias, privacy, dual-use, and alignment.","description":"# AI Ethics & Trade-offs Skill\n\nGenerate structured, nuanced analysis of AI safety, ethics, and capability trade-off questions — increasingly central to product decisions at AI companies.\n\n## When to Use\n- User asks about AI safety trade-offs\n- User asks about content policy decisions\n- User asks \"How would you handle [ethical dilemma in AI]?\"\n- User says `/ai-ethics-tradeoffs` followed by a question\n- Any question about responsible AI development, deployment, or governance\n- Especially relevant for safety-focused AI companies\n\n## Why This Matters\n\n**At safety-focused labs**: Safety is core to the company's identity. Every PM must reason about safety-capability trade-offs fluently.\n\n**At capability-focused labs**: Under increasing scrutiny for safety practices. PMs must articulate how to \"move fast\" responsibly.\n\n**At research-first labs**: Deep commitment to responsible AI development. PMs bridge research safety work and product decisions.\n\nThese questions increasingly distinguish good product thinking from great at any frontier AI company.\n\n## Framework: SAFE Method (5 Sections)\n\n### Section 1: Scope the Dilemma\nBefore analyzing, clearly define:\n- **The tension**: What two (or more) values are in conflict?\n- **The stakeholders**: Who is affected? (users, society, the company, specific communities, future generations)\n- **The timeframe**: Short-term vs. long-term implications\n- **The reversibility**: Can this decision be undone if wrong?\n\nCommon tension patterns in AI:\n- Safety vs. Capability (restricting model vs. making it more useful)\n- Access vs. Control (open-source vs. closed, free vs. gated)\n- Privacy vs. Personalization (user data vs. better experience)\n- Speed vs. Caution (shipping fast vs. thorough safety testing)\n- Transparency vs. Security (model details public vs. preventing misuse)\n\n### Section 2: Analyze Perspectives\nFor each stakeholder, articulate their legitimate concerns:\n- **Users**: What do they want? What risks do they face?\n- **Society**: What are the broader implications?\n- **Developers**: How does this affect those building on the platform?\n- **Researchers**: What does the scientific community need?\n- **Regulators**: What are the legal/compliance requirements?\n- **The company**: What are the business and reputational stakes?\n\n**Do NOT strawman any perspective.** The best answers demonstrate you can hold multiple valid viewpoints simultaneously.\n\n### Section 3: Framework Application\nApply one or more ethical frameworks:\n\n**Consequentialism**: What action produces the best outcome for the most people?\n- Expected value calculation (probability of harm x severity)\n- Short-term vs. long-term consequences\n- Direct vs. indirect effects\n\n**Deontological**: What are our obligations regardless of outcome?\n- User rights (privacy, autonomy, informed consent)\n- Company commitments (terms of service, safety pledges)\n- Professional ethics (do no harm, transparency)\n\n**Virtue Ethics**: What would a responsible AI company do?\n- Intellectual honesty (acknowledge uncertainty)\n- Precautionary principle (when in doubt, err on safety)\n- Proportionality (response matches the risk level)\n\n### Section 4: Evaluate Options\nPresent 3 approaches (spectrum from cautious to permissive):\n\n**Option A: Conservative / Safety-First**\n- What it looks like in practice\n- What you gain (safety, trust, regulatory goodwill)\n- What you lose (capability, user value, competitive position)\n\n**Option B: Balanced / Nuanced**\n- What it looks like in practice\n- How it threads the needle\n- What monitoring/adjustment mechanisms exist\n\n**Option C: Permissive / Capability-First**\n- What it looks like in practice\n- What you gain (innovation, user value, market position)\n- What you risk (harm, reputation, regulatory action)\n\n### Section 5: Recommend & Monitor\n- **Recommendation**: Pick an approach with clear reasoning\n- **Implementation**: How to execute it in practice\n- **Monitoring**: What signals would indicate it's working/failing\n- **Escalation criteria**: When would you revisit the decision?\n- **Communication**: How to explain this decision to different audiences\n\n## Key AI Ethics Topics\n\n### Content Policy & Moderation\n- Where to draw the line on model outputs\n- False positive refusals vs. harmful content getting through\n- Cultural context and global deployment\n- User expectations vs. safety requirements\n\n### Responsible Scaling\n- Responsible Scaling Policies (RSPs)\n- Preparedness Frameworks for frontier model deployment\n- Frontier AI safety approaches across major labs\n- When to slow down or stop scaling\n- Eval-gated deployment (capability thresholds that trigger safety reviews)\n\n### Bias & Fairness\n- Model bias in outputs (stereotyping, underrepresentation)\n- Training data bias and mitigation\n- Fairness across languages, cultures, and demographics\n- The tension between \"helpful\" and \"harmless\"\n\n### Privacy & Data\n- Training on user data (opt-in vs. opt-out)\n- Conversation privacy and data retention\n- Enterprise data isolation guarantees\n- Right to deletion and data portability\n\n### Dual-Use Concerns\n- Models that can help with both beneficial and harmful tasks\n- Biosecurity, cybersecurity, and weaponization risks\n- The \"publish or perish\" dilemma in AI research\n- Information hazards and responsible disclosure\n\n### Alignment & Control\n- How to ensure AI systems do what we intend\n- The principal-agent problem with AI assistants\n- Sycophancy vs. honest disagreement\n- When AI should refuse instructions\n\n### Economic Impact\n- Job displacement and workforce transition\n- Concentration of AI power in a few companies\n- Pricing and access (who gets to use AI?)\n- Impact on creative professions\n\n## Output Format\nWrite as a thoughtful, balanced analysis — not a sermon. Show you can reason about multiple perspectives without being paralyzed by them. Be opinionated but humble. Aim for ~2000 words.\n\n## Research-First Workflow\n1. **Research** — Search for recent incidents, policy decisions, research papers, and thought leader perspectives on the specific topic. Do 5-10 searches.\n2. **Cite sources** — Include `[linked source](url)` inline, especially for specific policies and incidents.\n3. **Display** the complete analysis.\n\n## What Good Looks Like\n- Identifies the core tension clearly (doesn't oversimplify)\n- Articulates multiple perspectives genuinely (not strawmanning)\n- Applies structured reasoning (not just gut feelings)\n- Makes a recommendation with conviction AND humility\n- Shows awareness of real-world examples and precedents\n- Connects ethical reasoning to product decisions (not just philosophy)\n- Demonstrates awareness of company-specific safety values and approaches","tags":["ethics","tradeoffs","skills","aroyburman-codes","agent-skills","claude-code","claude-skills","frameworks","metrics","pm-tools","product-management","product-strategy"],"capabilities":["skill","source-aroyburman-codes","skill-ai-ethics-tradeoffs","topic-agent-skills","topic-claude-code","topic-claude-skills","topic-frameworks","topic-metrics","topic-pm-tools","topic-product-management","topic-product-strategy"],"categories":["pm-skills"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/aroyburman-codes/pm-skills/ai-ethics-tradeoffs","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add aroyburman-codes/pm-skills","source_repo":"https://github.com/aroyburman-codes/pm-skills","install_from":"skills.sh"}},"qualityScore":"0.453","qualityRationale":"deterministic score 0.45 from registry signals: · indexed on github topic:agent-skills · 6 github stars · SKILL.md body (6,664 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T19:14:46.775Z","embedding":null,"createdAt":"2026-05-18T13:22:15.828Z","updatedAt":"2026-05-18T19:14:46.775Z","lastSeenAt":"2026-05-18T19:14:46.775Z","tsv":"'-10':858 '/ai-ethics-tradeoffs':86 '1':194,838 '2':297,860 '2000':832 '3':372,471,874 '4':467 '5':191,552,857 'access':259,793 'acknowledg':450 'across':643,677 'action':383,550 'affect':216,327 'agent':762 'ai':2,8,29,40,55,63,83,95,106,164,186,248,445,595,640,741,753,765,772,785,798 'ai-ethics-tradeoff':1 'aim':830 'align':28,748 'analysi':38,810,878 'analyz':199,298 'answer':362 'appli':375,897 'applic':374 'approach':472,558,642,938 'articul':149,303,891 'ask':61,69,75 'assist':766 'audienc':593 'autonomi':423 'awar':912,930 'b':506 'balanc':507,809 'benefici':726 'best':361,386 'better':276 'bias':22,663,666,673 'biosecur':730 'bridg':167 'broader':321 'build':329 'busi':351 'c':525 'calcul':394 'capability-first':527 'capability-focus':137 'capabl':12,44,131,138,251,500,528,657 'caution':280 'cautious':475 'central':50 'cite':861 'clear':200,560,887 'close':266 'commit':161,427 'common':244 'communic':585 'communiti':222,338 'compani':56,107,121,187,220,347,426,446,790,933 'company-specif':932 'competit':503 'complet':877 'concentr':783 'concern':306,719 'conflict':211 'connect':920 'consent':425 'consequ':407 'consequenti':381 'conserv':480 'content':20,71,598,614 'context':618 'control':261,749 'convers':701 'convict':908 'core':118,885 'cover':17 'creativ':801 'criteria':578 'cultur':617,679 'cybersecur':731 'data':274,672,689,693,704,707,714 'decis':53,73,173,239,584,590,845,925 'deep':160 'defin':201 'delet':712 'demograph':681 'demonstr':363,929 'deontolog':412 'deploy':97,621,638,656 'detail':291 'develop':96,165,323 'differ':592 'dilemma':81,197,739 'direct':408 'disagr':770 'disclosur':747 'discuss':16 'displac':779 'display':875 'distinguish':177 'doesn':888 'doubt':456 'draw':603 'dual':25,717 'dual-us':24,716 'econom':776 'effect':411 'ensur':752 'enterpris':706 'err':457 'escal':577 'especi':100,868 'ethic':3,10,30,42,80,379,434,440,596,921 'eval':654 'eval-g':653 'evalu':468 'everi':124 'exampl':917 'execut':565 'exist':523 'expect':392,623 'experi':277 'explain':588 'face':316 'fair':664,676 'fals':609 'fast':153,282 'feel':903 'first':158,483,529,836 'fluentli':135 'focus':105,114,139 'follow':87 'format':804 'framework':5,188,373,380,634 'free':267 'frontier':185,636,639 'futur':223 'gain':492,538 'gate':269,655 'generat':35,224 'genuin':894 'get':615,795 'global':620 'good':178,880 'goodwil':496 'govern':99 'great':182 'guarante':709 'gut':902 'handl':79 'harm':397,437,547,613,728 'harmless':687 'hazard':744 'help':685,723 'hold':366 'honest':769 'honesti':449 'humbl':829 'humil':910 'ident':123 'identifi':883 'impact':777,799 'implement':562 'implic':234,322 'incid':843,873 'includ':863 'increas':49,142,176 'indic':573 'indirect':410 'inform':424,743 'inlin':867 'innov':539 'instruct':775 'intellectu':448 'intend':758 'isol':708 'job':778 'key':594 'lab':115,140,159,645 'languag':678 'leader':850 'legal/compliance':344 'legitim':305 'level':465 'like':487,512,533,882 'line':605 'link':864 'long':232,405 'long-term':231,404 'look':486,511,532,881 'lose':499 'major':644 'make':255,904 'market':542 'match':462 'matter':110 'mechan':522 'method':190 'misus':295 'mitig':675 'model':253,290,607,637,665,720 'moder':600 'monitor':554,569 'monitoring/adjustment':521 'move':152 'multipl':367,819,892 'must':126,148 'navig':7 'need':339 'needl':519 'nuanc':37,508 'oblig':416 'off':33,67,134 'one':376 'open':263 'open-sourc':262 'opinion':827 'opt':695,699 'opt-in':694 'opt-out':698 'option':469,478,505,524 'outcom':387,419 'output':608,668,803 'oversimplifi':890 'paper':847 'paralyz':823 'pattern':246 'peopl':391 'perish':738 'permiss':477,526 'person':272 'perspect':299,359,820,851,893 'philosophi':928 'pick':556 'platform':332 'pledg':432 'pm':125 'pms':147,166 'polici':21,72,599,631,844,871 'portabl':715 'posit':504,543,610 'power':786 'practic':146,489,514,535,568 'precautionari':452 'preced':919 'prepared':633 'present':470 'prevent':294 'price':791 'princip':761 'principal-ag':760 'principl':453 'privaci':23,270,422,688,702 'probabl':395 'problem':763 'produc':384 'product':52,172,179,924 'profess':802 'profession':433 'proport':460 'public':292 'publish':736 'question':48,90,92,175 'real':915 'real-world':914 'reason':127,561,817,899,922 'recent':842 'recommend':553,555,906 'refus':611,774 'regardless':417 'regul':340 'regulatori':495,549 'relev':101 'reput':353,548 'requir':345,626 'research':157,168,333,742,835,839,846 'research-first':156,834 'respons':18,94,154,163,444,461,627,629,746 'restrict':252 'retent':705 'revers':236 'review':662 'revisit':582 'right':421,710 'risk':313,464,546,734 'rsps':632 'safe':189 'safeti':9,41,64,104,113,116,130,145,169,249,285,431,459,482,493,625,641,661,935 'safety-cap':129 'safety-first':481 'safety-focus':103,112 'say':85 'scale':19,628,630,652 'scientif':337 'scope':195 'scrutini':143 'search':840,859 'section':192,193,296,371,466,551 'secur':289 'sermon':813 'servic':430 'sever':399 'ship':281 'short':228,401 'short-term':227,400 'show':814,911 'signal':571 'simultan':370 'skill':34 'skill-ai-ethics-tradeoffs' 'slow':648 'societi':218,317 'sourc':264,862,865 'source-aroyburman-codes' 'specif':221,854,870,934 'spectrum':473 'speed':278 'stake':354 'stakehold':213,302 'stereotyp':669 'stop':651 'strawman':357,896 'structur':36,898 'sycoph':767 'system':754 'task':729 'tension':203,245,683,886 'term':229,233,402,406,428 'test':286 'think':180 'thorough':284 'thought':808,849 'thread':517 'threshold':658 'timefram':226 'topic':597,855 'topic-agent-skills' 'topic-claude-code' 'topic-claude-skills' 'topic-frameworks' 'topic-metrics' 'topic-pm-tools' 'topic-product-management' 'topic-product-strategy' 'trade':14,32,46,66,133 'trade-off':13,31,45,65,132 'tradeoff':4 'train':671,690 'transit':782 'transpar':287,438 'trigger':660 'trust':494 'two':205 'uncertainti':451 'underrepresent':670 'undon':241 'url':866 'use':26,59,258,718,797 'user':60,68,74,84,217,273,307,420,501,540,622,692 'valid':368 'valu':208,393,502,541,936 'viewpoint':369 'virtu':439 'vs':230,250,254,260,265,268,271,275,279,283,288,293,403,409,612,624,697,768 'want':311 'weapon':733 'without':821 'word':833 'work':170 'workflow':837 'workforc':781 'working/failing':576 'world':916 'would':77,442,572,580 'write':805 'wrong':243 'x':398","prices":[{"id":"59878853-2638-44be-9fba-f135b1bc9636","listingId":"d4edbfb2-32b2-48d5-a0e1-5b484a604a96","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"aroyburman-codes","category":"pm-skills","install_from":"skills.sh"},"createdAt":"2026-05-18T13:22:15.828Z"}],"sources":[{"listingId":"d4edbfb2-32b2-48d5-a0e1-5b484a604a96","source":"github","sourceId":"aroyburman-codes/pm-skills/ai-ethics-tradeoffs","sourceUrl":"https://github.com/aroyburman-codes/pm-skills/tree/main/skills/ai-ethics-tradeoffs","isPrimary":false,"firstSeenAt":"2026-05-18T13:22:15.828Z","lastSeenAt":"2026-05-18T19:14:46.775Z"}],"details":{"listingId":"d4edbfb2-32b2-48d5-a0e1-5b484a604a96","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"aroyburman-codes","slug":"ai-ethics-tradeoffs","github":{"repo":"aroyburman-codes/pm-skills","stars":6,"topics":["agent-skills","ai","claude-code","claude-skills","frameworks","metrics","pm-tools","product-management","product-strategy"],"license":"mit","html_url":"https://github.com/aroyburman-codes/pm-skills","pushed_at":"2026-02-17T06:52:03Z","description":"PM workflow and product thinking skills for AI product managers. 17 structured frameworks for PRDs, metrics, strategy, writing, prioritization, and more.","skill_md_sha":"2f056c300a6bdeaf8e0fd7d9e5afd940e2596fd7","skill_md_path":"skills/ai-ethics-tradeoffs/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/aroyburman-codes/pm-skills/tree/main/skills/ai-ethics-tradeoffs"},"layout":"multi","source":"github","category":"pm-skills","frontmatter":{"name":"ai-ethics-tradeoffs","description":"Framework for navigating AI safety, ethics, and capability trade-off discussions. Covers responsible scaling, content policy, bias, privacy, dual-use, and alignment."},"skills_sh_url":"https://skills.sh/aroyburman-codes/pm-skills/ai-ethics-tradeoffs"},"updatedAt":"2026-05-18T19:14:46.775Z"}}