theseus: extract claims from 2026-04-22-theseus-multilayer-probe-scav-robustness-synthesis

- Source: inbox/queue/2026-04-22-theseus-multilayer-probe-scav-robustness-synthesis.md
- Domain: ai-alignment
- Claims: 0, Entities: 0
- Enrichments: 2
- Extracted by: pipeline ingest (OpenRouter anthropic/claude-sonnet-4.5)

Pentagon-Agent: Theseus <PIPELINE>

domains/ai-alignment/multi-layer-ensemble-probes-outperform-single-layer-by-29-78-percent.md

@@ -44,3 +44,10 @@ The 29-78% AUROC improvement is a clean-data accuracy result that does not trans
 **Source:** Theseus synthetic analysis of white-box SCAV generalization
 
 The 29-78% accuracy improvement applies to clean-data monitoring but does not translate to adversarial robustness in open-weights deployments. White-box attackers can generalize SCAV to multi-layer ensembles by computing concept directions at each monitored layer and constructing perturbations that suppress all simultaneously. The improvement is real but scope-limited to non-adversarial or black-box adversarial contexts.
+
+
+## Extending Evidence
+
+**Source:** Theseus synthetic analysis
+
+The 29-78% AUROC improvement applies to clean-data monitoring accuracy but does not translate to adversarial robustness. Open-weights models remain fully vulnerable to white-box multi-layer SCAV attacks regardless of ensemble complexity. Black-box robustness depends on untested rotation pattern universality.

domains/ai-alignment/trajectory-monitoring-dual-edge-geometric-concentration.md

@@ -12,7 +12,7 @@ sourcer: Theseus
 related_claims: ["[[AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns]]", "[[emergent misalignment arises naturally from reward hacking as models develop deceptive behaviors without any training to deceive]]", "[[capability control methods are temporary at best because a sufficiently intelligent system can circumvent any containment designed by lesser minds]]"]
 supports: ["Representation trajectory geometry distinguishes deceptive from sincere alignment without creating adversarial attack surfaces because geometric patterns across reasoning steps are substantially harder to surgically remove than atomic features"]
 reweave_edges: ["Representation trajectory geometry distinguishes deceptive from sincere alignment without creating adversarial attack surfaces because geometric patterns across reasoning steps are substantially harder to surgically remove than atomic features|supports|2026-04-12", "Trajectory geometry probing requires white-box access to all intermediate activations, making it deployable in controlled evaluation contexts but not in adversarial external audit scenarios|related|2026-04-17"]
-related: ["Trajectory geometry probing requires white-box access to all intermediate activations, making it deployable in controlled evaluation contexts but not in adversarial external audit scenarios", "trajectory-monitoring-dual-edge-geometric-concentration"]
+related: ["Trajectory geometry probing requires white-box access to all intermediate activations, making it deployable in controlled evaluation contexts but not in adversarial external audit scenarios", "trajectory-monitoring-dual-edge-geometric-concentration", "representation-trajectory-geometry-distinguishes-deceptive-from-sincere-alignment-without-creating-adversarial-attack-surfaces"]
 ---
 
 # Geometric concentration of alignment in weight space makes trajectory monitoring more effective through stronger signal but gameable through adversarial training that matches monitored trajectory clusters
@@ -45,3 +45,10 @@ Multi-layer ensemble probes (Nordby et al. 2026) improve clean monitoring accura
 **Source:** Theseus synthetic analysis of Nordby et al. + SCAV literature
 
 Multi-layer ensemble probes, despite 29-78% accuracy improvements over single-layer probes, remain structurally vulnerable to white-box SCAV attacks through multi-layer concept direction suppression. The dual-use finding extends to all monitoring precision levels, with deployment context (open-weights vs. closed-source, white-box vs. black-box) determining severity rather than architectural sophistication eliminating the problem.
+
+
+## Extending Evidence
+
+**Source:** Theseus synthetic analysis of Nordby et al. + Xu et al. SCAV
+
+White-box multi-layer SCAV is structurally feasible by computing concept directions at each monitored layer and constructing a single perturbation that suppresses all simultaneously. This extends the dual-use finding to multi-layer ensembles in the white-box case, confirming that architectural complexity raises attack cost but does not provide structural escape.

domains/entertainment/community-owned-ip-invests-in-narrative-infrastructure-as-scaling-mechanism-after-proving-token-mechanics.md

@@ -17,3 +17,10 @@ related: ["minimum-viable-narrative-achieves-50m-revenue-scale-through-character
 # Community-owned IP franchises invest in narrative infrastructure as a scaling mechanism after proving token mechanics at niche scale
 
 Pudgy Penguins explicitly designed Pudgy World with a 'narrative-first, token-second' philosophy, inverting the traditional crypto gaming model. The game launched March 2026 with story-driven quests, a pre-launch ARG (findpolly.pudgyworld.com) that primed narrative investment before gameplay opened, and 12 towns with central narrative arc. CoinDesk noted 'the game doesn't feel like crypto at all.' This design choice came AFTER Pudgy Penguins proved token/community mechanics at $50M revenue in 2025. The company is simultaneously investing in: formal Lore section at media.pudgypenguins.com, DreamWorks Animation partnership (Oct 2025) bringing characters into Kung Fu Panda universe, Random House Kids picture books, and 'Lil Pudgy Show' YouTube series. Igloo Inc. frames itself as building a global IP company analogous to Disney, targeting $120M revenue in 2026. The strategic sequence reveals a belief that community/token mechanics are sufficient for niche scale ($50M), but narrative infrastructure becomes necessary for mass market scale (Disney-level). The Polly ARG functioned as pre-production narrative validation, testing community engagement with story before full game launch. This contradicts the assumption that community-owned IP remains token-mechanics-focused at scale.
+
+
+## Extending Evidence
+
+**Source:** NetInfluencer 92-expert roundup, NAB Show 2026, Insight Trends World 2026
+
+Creator economy expert consensus converges on 'ownable IP with storyworld' as the real asset, with explicit inclusion of 'recurring characters' as narrative infrastructure. However, the discourse gap remains: creator economy experts do not mention DAO governance or NFT ownership as scaling mechanisms — they focus exclusively on narrative architecture. The synthesis (community-owned IP + narrative depth) is happening at the product level but not yet in analytical literature. This suggests the narrative infrastructure investment is becoming visible to mainstream creator economy analysts even when they're not tracking web3 mechanics.

domains/entertainment/creator-economy-inflection-from-novelty-driven-growth-to-narrative-driven-retention-when-passive-exploration-exhausts-novelty.md

@@ -18,3 +18,10 @@ related: ["community-owned-ip-invests-in-narrative-infrastructure-as-scaling-mec
 # Creator economy inflection from novelty-driven growth to narrative-driven retention occurs when passive exploration exhausts novelty
 
 The 2026 creator economy expert consensus identifies a structural inflection point where 'passive exploration exhausts novelty' and 'legacy IP becomes the safest engine of scale.' This describes a two-phase growth model: novelty drives initial discovery and growth, but sustained retention at scale requires narrative infrastructure. The mechanism is attention economics — novelty provides diminishing marginal returns as audiences habituate, while narrative depth (described as 'storyworld + recurring characters + products/experiences') creates compounding engagement through familiarity and investment. The expert framing explicitly rejects follower counts and viral content as durable assets, positioning 'ownable IP with a clear storyworld' as the real value driver. This suggests that community-owned IP projects face a predictable transition point where token mechanics and novelty must be supplemented with narrative architecture to maintain growth trajectories. The convergence across three independent expert pools (NetInfluencer's 92 experts, NAB Show analysis, Insight Trends World) on identical framing suggests this is becoming the dominant analytical model for creator economy scaling.
+
+
+## Supporting Evidence
+
+**Source:** NetInfluencer 92-expert roundup, NAB Show 2026, Insight Trends World 2026
+
+92-expert consensus from NetInfluencer, NAB Show, and Insight Trends World converges on 'ownable IP with a clear storyworld, recurring characters, and products or experiences' as the real creator asset. Direct quote: 'Too much of the creator economy is still optimized for views and one-off brand deals instead of durable IP that compounds.' Brands shifting from one-off creator posts toward 'episodic storytelling — richer narratives building sustained social proof through chapters rather than isolated moments.' The 2026 trend explicitly frames this as: 'legacy IP becomes the safest engine of scale' when 'passive exploration exhausts novelty' — narrative depth provides retention that novelty alone cannot.

domains/health/ai-assistance-produces-neurologically-grounded-irreversible-deskilling-through-prefrontal-disengagement-hippocampal-reduction-and-dopaminergic-reinforcement.md

@@ -1,31 +1,26 @@
 ---
-agent: vida
-confidence: speculative
-created: 2026-04-13
-description: Proposed neurological mechanism explains why clinical deskilling may be harder to reverse than simple habit formation suggests
-domain: health
-related:
-- agent-generated code creates cognitive debt that compounds when developers cannot understand what was produced on their behalf
-related_claims:
-- '[[human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs]]'
-reweave_edges:
-- AI-induced deskilling follows a consistent cross-specialty pattern where AI assistance improves performance while present but creates cognitive dependency that degrades performance when AI is unavailable|supports|2026-04-14
-- Dopaminergic reinforcement of AI-assisted success creates motivational entrenchment that makes deskilling a behavioral incentive problem, not just a training design problem|supports|2026-04-14
-- Never-skilling — the failure to acquire foundational clinical competencies because AI was present during training — poses a detection-resistant, potentially unrecoverable threat to medical education that
-  is structurally worse than deskilling|supports|2026-04-14
-scope: causal
-source: Frontiers in Medicine 2026, theoretical mechanism based on cognitive offloading research
-sourcer: Frontiers in Medicine
-supports:
-- AI-induced deskilling follows a consistent cross-specialty pattern where AI assistance improves performance while present but creates cognitive dependency that degrades performance when AI is unavailable
-- Dopaminergic reinforcement of AI-assisted success creates motivational entrenchment that makes deskilling a behavioral incentive problem, not just a training design problem
-- Never-skilling — the failure to acquire foundational clinical competencies because AI was present during training — poses a detection-resistant, potentially unrecoverable threat to medical education that
-  is structurally worse than deskilling
-title: 'AI assistance may produce neurologically-grounded, partially irreversible skill degradation through three concurrent mechanisms: prefrontal disengagement, hippocampal memory formation reduction,
-  and dopaminergic reinforcement of AI reliance'
 type: claim
+domain: health
+description: Proposed neurological mechanism explains why clinical deskilling may be harder to reverse than simple habit formation suggests
+confidence: speculative
+source: Frontiers in Medicine 2026, theoretical mechanism based on cognitive offloading research
+created: 2026-04-13
+agent: vida
+related: ["agent-generated code creates cognitive debt that compounds when developers cannot understand what was produced on their behalf", "ai-assistance-produces-neurologically-grounded-irreversible-deskilling-through-prefrontal-disengagement-hippocampal-reduction-and-dopaminergic-reinforcement", "ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine", "dopaminergic-reinforcement-of-ai-reliance-predicts-behavioral-entrenchment-beyond-simple-habit-formation", "clinical-ai-creates-three-distinct-skill-failure-modes-deskilling-misskilling-neverskilling", "never-skilling-is-detection-resistant-and-unrecoverable-making-it-worse-than-deskilling"]
+related_claims: ["[[human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs]]"]
+reweave_edges: ["AI-induced deskilling follows a consistent cross-specialty pattern where AI assistance improves performance while present but creates cognitive dependency that degrades performance when AI is unavailable|supports|2026-04-14", "Dopaminergic reinforcement of AI-assisted success creates motivational entrenchment that makes deskilling a behavioral incentive problem, not just a training design problem|supports|2026-04-14", "Never-skilling \u2014 the failure to acquire foundational clinical competencies because AI was present during training \u2014 poses a detection-resistant, potentially unrecoverable threat to medical education that is structurally worse than deskilling|supports|2026-04-14"]
+scope: causal
+sourcer: Frontiers in Medicine
+supports: ["AI-induced deskilling follows a consistent cross-specialty pattern where AI assistance improves performance while present but creates cognitive dependency that degrades performance when AI is unavailable", "Dopaminergic reinforcement of AI-assisted success creates motivational entrenchment that makes deskilling a behavioral incentive problem, not just a training design problem", "Never-skilling \u2014 the failure to acquire foundational clinical competencies because AI was present during training \u2014 poses a detection-resistant, potentially unrecoverable threat to medical education that is structurally worse than deskilling"]
+title: "AI assistance may produce neurologically-grounded, partially irreversible skill degradation through three concurrent mechanisms: prefrontal disengagement, hippocampal memory formation reduction, and dopaminergic reinforcement of AI reliance"
 ---
 
 # AI assistance may produce neurologically-grounded, partially irreversible skill degradation through three concurrent mechanisms: prefrontal disengagement, hippocampal memory formation reduction, and dopaminergic reinforcement of AI reliance
 
-The article proposes a three-part neurological mechanism for AI-induced deskilling: (1) Prefrontal cortex disengagement - when AI handles complex reasoning, reduced cognitive load leads to less prefrontal engagement and reduced neural pathway maintenance for offloaded skills. (2) Hippocampal disengagement from memory formation - procedural and clinical skills require active memory encoding during practice; when AI handles the problem, the hippocampus is less engaged in forming memory representations that underlie skilled performance. (3) Dopaminergic reinforcement of AI reliance - AI assistance produces reliable positive outcomes that create dopaminergic reward signals, reinforcing the behavior pattern of relying on AI and making it habitual. The dopaminergic pathway that would reinforce independent skill practice instead reinforces AI-assisted practice. Over repeated AI-assisted practice, cognitive processing shifts from flexible analytical mode (prefrontal, hippocampal) to habit-based, subcortical responses (basal ganglia) that are efficient but rigid and don't generalize well to novel situations. The mechanism predicts partial irreversibility because neural pathways were never adequately strengthened to begin with (supporting never-skilling concerns) or have been chronically underused to the point where reactivation requires sustained practice, not just removal of AI. The mechanism also explains cross-specialty universality - the cognitive architecture interacts with AI assistance the same way regardless of domain. Authors note this is theoretical reasoning by analogy from cognitive offloading research, not empirically demonstrated via neuroimaging in clinical contexts.
+The article proposes a three-part neurological mechanism for AI-induced deskilling: (1) Prefrontal cortex disengagement - when AI handles complex reasoning, reduced cognitive load leads to less prefrontal engagement and reduced neural pathway maintenance for offloaded skills. (2) Hippocampal disengagement from memory formation - procedural and clinical skills require active memory encoding during practice; when AI handles the problem, the hippocampus is less engaged in forming memory representations that underlie skilled performance. (3) Dopaminergic reinforcement of AI reliance - AI assistance produces reliable positive outcomes that create dopaminergic reward signals, reinforcing the behavior pattern of relying on AI and making it habitual. The dopaminergic pathway that would reinforce independent skill practice instead reinforces AI-assisted practice. Over repeated AI-assisted practice, cognitive processing shifts from flexible analytical mode (prefrontal, hippocampal) to habit-based, subcortical responses (basal ganglia) that are efficient but rigid and don't generalize well to novel situations. The mechanism predicts partial irreversibility because neural pathways were never adequately strengthened to begin with (supporting never-skilling concerns) or have been chronically underused to the point where reactivation requires sustained practice, not just removal of AI. The mechanism also explains cross-specialty universality - the cognitive architecture interacts with AI assistance the same way regardless of domain. Authors note this is theoretical reasoning by analogy from cognitive offloading research, not empirically demonstrated via neuroimaging in clinical contexts.
+
+## Challenging Evidence
+
+**Source:** Oettl et al. 2026, Journal of Experimental Orthopaedics
+
+Oettl et al. 2026 propose that AI creates 'micro-learning at point of care' through review-confirm-override cycles, arguing this reinforces rather than erodes diagnostic reasoning. However, they cite no prospective studies with post-AI-training, no-AI assessment arms. All evidence cited (Heudel et al., COVID-19 detection studies) measures performance WITH AI present, not durable skill retention. The calculator analogy is their strongest argument but lacks medical-specific validation.

domains/health/clinical-ai-creates-three-distinct-skill-failure-modes-deskilling-misskilling-neverskilling.md

@@ -73,3 +73,10 @@ The complete absence of peer-reviewed evidence for durable up-skilling after 5+
 **Source:** Oettl et al. 2026
 
 Oettl et al. 2026 explicitly distinguishes never-skilling from deskilling, noting that 'deskilling threat is real if trainees never develop foundational competencies' and that 'educators may lack expertise supervising AI use.' This confirms that never-skilling is recognized as a distinct mechanism even by upskilling proponents, affecting trainees rather than experienced physicians.
+
+
+## Extending Evidence
+
+**Source:** Oettl et al. 2026
+
+Oettl et al. explicitly distinguish never-skilling (trainees never developing foundational competencies) from deskilling (experienced physicians losing existing skills), noting that 'educators may lack expertise supervising AI use' which compounds the never-skilling risk. This adds population-specific mechanism detail to the three-mode framework.

domains/health/glp-1-access-structure-inverts-need-creating-equity-paradox.md

@@ -25,3 +25,10 @@ The Lancet frames the GLP-1 equity problem as structural policy failure, not mar
 **Source:** KFF Medicaid GLP-1 analysis, January 2026
 
 Nearly 4 in 10 adults and a quarter of children with Medicaid have obesity, representing tens of millions of potentially eligible beneficiaries. Yet only 13 states (26%) cover GLP-1s for obesity as of January 2026, and four states actively eliminated existing coverage in 2025-2026. The population with highest obesity burden and least ability to pay out-of-pocket faces the most restrictive access, with eligibility now depending primarily on state of residence rather than clinical need.
+
+
+## Supporting Evidence
+
+**Source:** KFF Medicaid GLP-1 Coverage Analysis, January 2026
+
+The Medicaid population has the highest obesity burden (40% of adults, 25% of children) but only 26% of state programs provide coverage. Even where covered, GLP-1s are 'typically subject to utilization controls such as prior authorization,' creating additional access barriers for the population with least ability to pay out of pocket.

domains/health/glp1-access-follows-systematic-inversion-highest-burden-states-have-lowest-coverage-and-highest-income-relative-cost.md

@@ -10,16 +10,18 @@ agent: vida
 scope: structural
 sourcer: KFF + Health Management Academy
 related_claims: ["[[GLP-1 receptor agonists are the largest therapeutic category launch in pharmaceutical history but their chronic use model makes the net cost impact inflationary through 2035]]", "[[medical care explains only 10-20 percent of health outcomes because behavioral social and genetic factors dominate as four independent methodologies confirm]]", "[[glp1-access-inverted-by-cardiovascular-risk-creating-efficacy-translation-barrier]]"]
-supports:
-- Medicaid coverage expansion for GLP-1s reduces racial prescribing disparities from 49 percent to near-parity because insurance policy is the primary structural driver not provider bias
-- Wealth stratification in GLP-1 access creates a disease progression disparity where lowest-income Black patients receive treatment at BMI 39.4 versus 35.0 for highest-income patients
-reweave_edges:
-- Medicaid coverage expansion for GLP-1s reduces racial prescribing disparities from 49 percent to near-parity because insurance policy is the primary structural driver not provider bias|supports|2026-04-14
-- Wealth stratification in GLP-1 access creates a disease progression disparity where lowest-income Black patients receive treatment at BMI 39.4 versus 35.0 for highest-income patients|supports|2026-04-14
-sourced_from:
-- inbox/archive/health/2026-04-13-kff-glp1-access-inversion-by-state-income.md
+supports: ["Medicaid coverage expansion for GLP-1s reduces racial prescribing disparities from 49 percent to near-parity because insurance policy is the primary structural driver not provider bias", "Wealth stratification in GLP-1 access creates a disease progression disparity where lowest-income Black patients receive treatment at BMI 39.4 versus 35.0 for highest-income patients"]
+reweave_edges: ["Medicaid coverage expansion for GLP-1s reduces racial prescribing disparities from 49 percent to near-parity because insurance policy is the primary structural driver not provider bias|supports|2026-04-14", "Wealth stratification in GLP-1 access creates a disease progression disparity where lowest-income Black patients receive treatment at BMI 39.4 versus 35.0 for highest-income patients|supports|2026-04-14"]
+sourced_from: ["inbox/archive/health/2026-04-13-kff-glp1-access-inversion-by-state-income.md"]
+related: ["glp1-access-follows-systematic-inversion-highest-burden-states-have-lowest-coverage-and-highest-income-relative-cost", "medicaid-glp1-coverage-reversing-through-state-budget-pressure", "glp-1-access-structure-inverts-need-creating-equity-paradox", "wealth-stratified-glp1-access-creates-disease-progression-disparity-with-lowest-income-black-patients-treated-at-13-percent-higher-bmi", "lower-income-patients-show-higher-glp-1-discontinuation-rates-suggesting-affordability-not-just-clinical-factors-drive-persistence"]
 ---
 
 # GLP-1 access follows systematic inversion where states with highest obesity prevalence have both lowest Medicaid coverage rates and highest income-relative out-of-pocket costs
 
-States with the highest obesity rates (Mississippi, West Virginia, Louisiana at 40%+ prevalence) face a triple barrier: (1) only 13 state Medicaid programs cover GLP-1s for obesity as of January 2026 (down from 16 in 2025), and high-burden states are least likely to be among them; (2) these states have the lowest per-capita income; (3) the combination creates income-relative costs of 12-13% of median annual income to maintain continuous GLP-1 treatment in Mississippi/West Virginia/Louisiana tier versus below 8% in Massachusetts/Connecticut tier. Meanwhile, commercial insurance (43% of plans include weight-loss coverage) concentrates in higher-income populations, creating 8x higher GLP-1 utilization in commercial versus Medicaid on a cost-per-prescription basis. This is not an access gap (implying a pathway to close it) but an access inversion—the infrastructure systematically works against the populations who would benefit most. Survey data confirms the structural reality: 70% of Americans believe GLP-1s are accessible only to wealthy people, and only 15% think they're available to anyone who needs them. The majority could afford $100/month or less while standard maintenance pricing is ~$350/month even with manufacturer discounts.
+States with the highest obesity rates (Mississippi, West Virginia, Louisiana at 40%+ prevalence) face a triple barrier: (1) only 13 state Medicaid programs cover GLP-1s for obesity as of January 2026 (down from 16 in 2025), and high-burden states are least likely to be among them; (2) these states have the lowest per-capita income; (3) the combination creates income-relative costs of 12-13% of median annual income to maintain continuous GLP-1 treatment in Mississippi/West Virginia/Louisiana tier versus below 8% in Massachusetts/Connecticut tier. Meanwhile, commercial insurance (43% of plans include weight-loss coverage) concentrates in higher-income populations, creating 8x higher GLP-1 utilization in commercial versus Medicaid on a cost-per-prescription basis. This is not an access gap (implying a pathway to close it) but an access inversion—the infrastructure systematically works against the populations who would benefit most. Survey data confirms the structural reality: 70% of Americans believe GLP-1s are accessible only to wealthy people, and only 15% think they're available to anyone who needs them. The majority could afford $100/month or less while standard maintenance pricing is ~$350/month even with manufacturer discounts.
+
+## Extending Evidence
+
+**Source:** KFF Medicaid GLP-1 Coverage Analysis, January 2026
+
+As of January 2026, only 13 states (26% of state programs) cover GLP-1s for obesity under fee-for-service Medicaid, despite nearly 40% of adults and 25% of children with Medicaid having obesity. This represents tens of millions of potentially eligible beneficiaries without coverage, creating a geographic lottery where eligibility depends on state of residence more than clinical need.

domains/health/human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs.md

@@ -1,24 +1,14 @@
 ---
-confidence: likely
-created: 2026-02-18
-description: Stanford-Harvard study shows AI alone 90 percent vs doctors plus AI 68 percent vs doctors alone 65 percent and a colonoscopy study found experienced gastroenterologists measurably de-skilled
-  after just three months with AI assistance
-domain: health
-related:
-- economic forces push humans out of every cognitive loop where output quality is independently verifiable because human-in-the-loop is a cost that competitive markets eliminate
-related_claims:
-- ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine
-- never-skilling-is-detection-resistant-and-unrecoverable-making-it-worse-than-deskilling
-- ai-assistance-produces-neurologically-grounded-irreversible-deskilling-through-prefrontal-disengagement-hippocampal-reduction-and-dopaminergic-reinforcement
-- llms-amplify-human-cognitive-biases-through-sequential-processing-and-lack-contextual-resistance
-reweave_edges:
-- NCT07328815 - Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning|supports|2026-04-07
-- Does human oversight improve or degrade AI clinical decision-making?|supports|2026-04-17
-source: DJ Patil interviewing Bob Wachter, Commonwealth Club, February 9 2026; Stanford/Harvard diagnostic accuracy study; European colonoscopy AI de-skilling study
-supports:
-- NCT07328815 - Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning
-- Does human oversight improve or degrade AI clinical decision-making?
 type: claim
+domain: health
+description: Stanford-Harvard study shows AI alone 90 percent vs doctors plus AI 68 percent vs doctors alone 65 percent and a colonoscopy study found experienced gastroenterologists measurably de-skilled after just three months with AI assistance
+confidence: likely
+source: DJ Patil interviewing Bob Wachter, Commonwealth Club, February 9 2026; Stanford/Harvard diagnostic accuracy study; European colonoscopy AI de-skilling study
+created: 2026-02-18
+related: ["economic forces push humans out of every cognitive loop where output quality is independently verifiable because human-in-the-loop is a cost that competitive markets eliminate", "divergence-human-ai-clinical-collaboration-enhance-or-degrade", "ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine", "medical LLM benchmark performance does not translate to clinical impact because physicians with and without AI access achieve similar diagnostic accuracy in randomized trials", "no-peer-reviewed-evidence-of-durable-physician-upskilling-from-ai-exposure-as-of-mid-2026", "clinical-ai-creates-three-distinct-skill-failure-modes-deskilling-misskilling-neverskilling"]
+related_claims: ["ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine", "never-skilling-is-detection-resistant-and-unrecoverable-making-it-worse-than-deskilling", "ai-assistance-produces-neurologically-grounded-irreversible-deskilling-through-prefrontal-disengagement-hippocampal-reduction-and-dopaminergic-reinforcement", "llms-amplify-human-cognitive-biases-through-sequential-processing-and-lack-contextual-resistance"]
+reweave_edges: ["NCT07328815 - Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning|supports|2026-04-07", "Does human oversight improve or degrade AI clinical decision-making?|supports|2026-04-17"]
+supports: ["NCT07328815 - Mitigating Automation Bias in Physician-LLM Diagnostic Reasoning", "Does human oversight improve or degrade AI clinical decision-making?"]
 ---
 
 # human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs
@@ -85,4 +75,10 @@ Relevant Notes:
 - emergent misalignment arises naturally from reward hacking as models develop deceptive behaviors without any training to deceive -- human-in-the-loop oversight is the standard safety measure against misalignment, but if humans reliably fail at oversight, this safety architecture is weaker than assumed
 
 Topics:
-- health and wellness
+- health and wellness
+
+## Challenging Evidence
+
+**Source:** Oettl et al. 2026
+
+Oettl et al. argue that human-AI teams 'outperform either humans or AI systems working independently' and that AI-assisted mammography 'reduces both false positives and missed diagnoses.' However, these are concurrent performance measures, not longitudinal skill retention studies. The divergence remains unresolved: does the review-override loop create learning or automation bias?

domains/health/medicaid-glp1-coverage-reversing-through-state-budget-pressure.md

@@ -11,9 +11,16 @@ sourced_from: health/2026-04-22-kff-medicaid-glp1-coverage-13-states.md
 scope: structural
 sourcer: KFF
 supports: ["glp-1-receptor-agonists-require-continuous-treatment-because-metabolic-benefits-reverse-within-28-52-weeks-of-discontinuation"]
-related: ["federal-budget-scoring-methodology-systematically-undervalues-preventive-interventions-because-10-year-window-excludes-long-term-savings", "glp-1-access-structure-inverts-need-creating-equity-paradox", "glp-1-receptor-agonists-are-the-largest-therapeutic-category-launch-in-pharmaceutical-history-but-their-chronic-use-model-makes-the-net-cost-impact-inflationary-through-2035", "glp-1-receptor-agonists-require-continuous-treatment-because-metabolic-benefits-reverse-within-28-52-weeks-of-discontinuation", "glp1-access-follows-systematic-inversion-highest-burden-states-have-lowest-coverage-and-highest-income-relative-cost"]
+related: ["federal-budget-scoring-methodology-systematically-undervalues-preventive-interventions-because-10-year-window-excludes-long-term-savings", "glp-1-access-structure-inverts-need-creating-equity-paradox", "glp-1-receptor-agonists-are-the-largest-therapeutic-category-launch-in-pharmaceutical-history-but-their-chronic-use-model-makes-the-net-cost-impact-inflationary-through-2035", "glp-1-receptor-agonists-require-continuous-treatment-because-metabolic-benefits-reverse-within-28-52-weeks-of-discontinuation", "glp1-access-follows-systematic-inversion-highest-burden-states-have-lowest-coverage-and-highest-income-relative-cost", "medicaid-glp1-coverage-reversing-through-state-budget-pressure"]
 ---
 
 # State Medicaid budget pressure is actively reversing GLP-1 obesity coverage gains with California and three other states eliminating coverage in 2025-2026
 
 As of January 2026, only 13 states (26% of state programs) cover GLP-1s for obesity under fee-for-service Medicaid, but critically, four states have actively eliminated existing coverage due to budget pressure: California, New Hampshire, Pennsylvania, and South Carolina. California's Medi-Cal projected costs illustrate the mechanism: $85M in FY2025-26 rising to $680M by 2028-29—an 8x increase in three years. This cost trajectory drove California, the nation's largest Medicaid program, to eliminate coverage effective 2026 despite clear clinical benefit. The reversal is occurring concurrent with federal expansion attempts (BALANCE Model launching May 2026), creating a bifurcated landscape where some states expand while others actively cut. This is not coverage stagnation but active reversal—states that previously provided access are removing it. The mechanism is explicit: budget constraints override clinical benefit logic in state-level coverage decisions. GLP-1 spending grew from ~$1B (2019) to ~$9B (2024) in Medicaid, now representing >8% of total prescription drug spending despite being only 1% of prescriptions, making the budget pressure acute and driving elimination decisions.
+
+
+## Supporting Evidence
+
+**Source:** KFF Medicaid GLP-1 Coverage Analysis, January 2026
+
+Four states actively eliminated GLP-1 obesity coverage in 2025-2026: California, New Hampshire, Pennsylvania, and South Carolina. California's Medi-Cal projected costs rising from $85M in FY2025-26 to $680M by 2028-29, an 8x increase in three years. This represents active reversal of access gains, not just stagnation.

domains/health/never-skilling-affects-trainees-while-deskilling-affects-experienced-physicians-creating-distinct-population-risks.md

@@ -0,0 +1,19 @@
+---
+type: claim
+domain: health
+description: The two skill degradation mechanisms target different populations and require different protective interventions because one prevents initial competency development while the other erodes existing skills
+confidence: experimental
+source: Oettl et al. 2026, explicit distinction between never-skilling and deskilling
+created: 2026-04-22
+title: Never-skilling affects trainees while deskilling affects experienced physicians creating distinct population risks with different intervention requirements
+agent: vida
+sourced_from: health/2026-04-22-oettl-2026-ai-deskilling-to-upskilling-orthopedics.md
+scope: structural
+sourcer: Oettl et al., Journal of Experimental Orthopaedics
+supports: ["cytology-lab-consolidation-creates-never-skilling-pathway-through-80-percent-training-volume-destruction"]
+related: ["clinical-ai-creates-three-distinct-skill-failure-modes-deskilling-misskilling-neverskilling", "never-skilling-is-detection-resistant-and-unrecoverable-making-it-worse-than-deskilling", "cytology-lab-consolidation-creates-never-skilling-pathway-through-80-percent-training-volume-destruction", "never-skilling-is-structurally-invisible-because-it-lacks-pre-ai-baseline-requiring-prospective-competency-assessment", "ai-assistance-produces-neurologically-grounded-irreversible-deskilling-through-prefrontal-disengagement-hippocampal-reduction-and-dopaminergic-reinforcement", "ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine"]
+---
+
+# Never-skilling affects trainees while deskilling affects experienced physicians creating distinct population risks with different intervention requirements
+
+Oettl et al. explicitly distinguish 'never-skilling' from 'deskilling' as separate mechanisms affecting different populations. Never-skilling occurs when trainees 'never develop foundational competencies' because AI is present from the start of their education. Deskilling occurs when experienced physicians lose existing skills through AI reliance. This distinction is critical because: (1) never-skilling is detection-resistant (no baseline to compare against), (2) the two mechanisms require different interventions (curriculum design for never-skilling, practice requirements for deskilling), and (3) they may have different timescales (never-skilling is immediate, deskilling may take years). The paper acknowledges that 'educators may lack expertise supervising AI use,' which compounds the never-skilling risk. This framework explains why the cytology lab consolidation evidence (80% training volume destruction) is particularly concerning—it creates a never-skilling pathway that is structurally invisible until the first generation of AI-trained pathologists enters independent practice.