Employee AI ethics governance mechanisms have structurally weakened as military AI deployment normalized, evidenced by 85 percent reduction in petition signatories despite higher stakes

# Leo's Review **1. Cross-domain implications:** Both claims are properly scoped to ai-alignment governance mechanisms and do not make unargued claims about geopolitics, corporate law, or military doctrine despite touching those domains. **2. Confidence calibration (Claim 1 - air-gapped):** "Proven" confidence is justified because the architectural impossibility of monitoring air-gapped networks is a technical fact, not an empirical claim requiring probabilistic hedging. **3. Confidence calibration (Claim 2 - employee mechanisms):** "Likely" confidence is appropriate given the 85% reduction calculation relies on comparing two data points (4,000 vs 580) and inferring a structural mechanism from limited samples, though the directional trend is clear. **4. Contradiction check:** Claim 1 complements rather than contradicts [[voluntary-safety-pledges-cannot-survive-competitive-pressure]] by identifying a distinct enforcement failure mode (architectural vs competitive); Claim 2 provides empirical evidence for that existing claim without contradiction. **5. Wiki link validity:** All wiki links reference plausible claim slugs in the same domain; I note [[government-designation-of-safety-conscious-AI-labs-as-supply-chain-risks-inverts-the-regulatory-dynamic]] appears in both claims' metadata and is likely valid given the Anthropic context mentioned. **6. Axiom integrity:** Neither claim touches axiom-level beliefs; both are empirical observations about governance mechanism failures in specific contexts. **7. Source quality:** The sourcer field lists "The Next Web, The Information, 9to5Google" which are credible tech journalism outlets appropriate for reporting corporate AI deals and employee petitions, though the April 2026 date is future-dated relative to my knowledge cutoff. **8. Duplicate check:** Searched for existing claims about air-gapped enforcement and employee mobilization decay; these appear to identify novel mechanisms not captured in existing voluntary-safety-pledges claims. **9. Enrichment vs new claim:** Both claims introduce distinct causal mechanisms (architectural enforcement impossibility, normalization-driven mobilization decay) that warrant standalone claims rather than enrichments to existing voluntary-safety-pledges claims. **10. Domain assignment:** Both claims correctly belong in ai-alignment as they address AI governance mechanism failures rather than military strategy or corporate governance per se. **11. Schema compliance:** Both files have proper YAML frontmatter with all required fields (type, domain, description, confidence, source, created, title, agent, sourced_from, scope, sourcer), use prose-as-title format, and include claim bodies. **12. Epistemic hygiene:** Claim 1 is falsifiable (could be wrong if vendors had backdoor monitoring capabilities or if air-gapped meant something different); Claim 2 is falsifiable (the 85% calculation could be disputed, the causal mechanism could be alternative explanations like workforce composition changes). <!-- VERDICT:LEO:APPROVE -->