{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"negative_space_design__tech_ethics_ai_governance","hypothesis_id":"H5","search_queries":["AI governance portfolio budget percentage reserved assurance testing evaluation capacity","algorithmic impact assessment dedicated budget percentage pre-deployment testing","AI regulatory sandbox resource allocation portfolio capacity reserve non deployment","responsible AI governance three lines model independent testing budget capacity","\"ring-fenced\" budget AI assurance evaluation red teaming","innovation portfolio \"10%\" capacity experimentation reserve","patent AI governance portfolio resource allocation red team evaluation capacity","ISO 42001 resource allocation independent AI evaluation pre deployment official"],"sources":[{"source_id":"C1","title":"Ethical AI Framework","publisher":"Digital Policy Office, Government of the Hong Kong Special Administrative Region","url":"https://www.digitalpolicy.gov.hk/en/our_work/data_governance/policies_standards/ethical_ai_framework/doc/Ethical_AI_Framework_en.pdf","source_class":"OFFICIAL_GUIDANCE","claims_supported":["The framework assigns independent review, challenge, acceptance criteria, and pre-delivery sign-off to governance functions separate from the AI project team.","It requires assessment and higher-level approval for high-risk AI applications but does not prescribe a fixed portfolio-capacity reserve."]},{"source_id":"C2","title":"Three lines of defense against risks from AI","publisher":"Centre for the Governance of AI / AI & Society","url":"https://cdn.governance.ai/Three_Lines_of_Defense_Against_Risks_From_AI.pdf","source_class":"PRIMARY_RESEARCH","claims_supported":["The paper describes dedicated second-line risk expertise and independent third-line assurance, including audits and red teaming before or after deployment.","It identifies lack of expertise, information, or time as a cause of risk-coverage gaps that can permit unsafe deployment."]},{"source_id":"C3","title":"Costed Evaluation Plan Guidance, Tools and Templates","publisher":"United Nations Population Fund","url":"https://www.unfpa.org/sites/default/files/admin-resource/CEPlan%20Guidance%2C%20Tools%20and%20Templates-307.pdf","source_class":"OFFICIAL_GUIDANCE","claims_supported":["UNFPA places evaluation funds inside an annual organizational resource ceiling and ring-fences them exclusively for evaluation work.","The funds cannot be used for other activities without formal approval, closely matching the proposed protection and release-boundary mechanism in a neighboring domain."]},{"source_id":"C4","title":"Innovation portfolios for public sector organizations","publisher":"Deloitte Insights","url":"https://www.deloitte.com/us/en/insights/industry/government-public-sector-services/innovation-portfolios-public-sector-organizations.html","source_class":"AUTHORITATIVE_SECONDARY","claims_supported":["Portfolio practice includes explicit percentage allocations such as 70-20-10 across core, adjacent, and transformational innovation.","Learning value can justify uncertain portfolio investments, but the allocated categories pursue innovation rather than protected harm inquiry."]},{"source_id":"C5","title":"Innovation Portfolio Management for the Public Non-Profit R&D Sector","publisher":"Cornell University Dyson School repository","url":"https://barrett.dyson.cornell.edu/files/papers/Manuscript_resubmit%20clean.pdf","source_class":"PRIMARY_RESEARCH","claims_supported":["The review treats formal resource-allocation strategy, experimentation, learning, and portfolio governance as established innovation-management components.","It reports that information gaps and financially narrow selection criteria can produce organizational lock-in, while clear allocation strategies can preserve learning-oriented work."]},{"source_id":"C6","title":"Anthropic’s Responsible Scaling Policy","publisher":"Anthropic","url":"https://www.anthropic.com/responsible-scaling-policy","source_class":"COMMERCIAL_FIRST_PARTY","claims_supported":["Anthropic documents recurring capability evaluations, risk reports, external review, red teaming, and circumstances permitting pauses in AI development.","The policy operationalizes protected safety work and deployment restraint but does not allocate a minimum share of an annual multi-pilot portfolio to nondeployment inquiry."]},{"source_id":"C7","title":"Portfolio of AI assurance techniques","publisher":"UK Department for Science, Innovation and Technology","url":"https://www.gov.uk/data-ethics-guidance/portfolio-of-ai-assurance-techniques","source_class":"OFFICIAL_GUIDANCE","claims_supported":["The official portfolio collects assurance techniques used across the AI lifecycle to support trustworthy development, deployment, and procurement.","It establishes assurance as a portfolio of practices, not as a ring-fenced percentage of organizational pilot capacity."]},{"source_id":"C8","title":"Systems and Methods of Facilitating Intelligent Budget Governance for a Cloud Resource","publisher":"U.S. Patent Application 20260057320, reproduced by Justia Patents","url":"https://patents.justia.com/patent/20260057320","source_class":"PRIMARY_RESEARCH","claims_supported":["The application describes annual and other planning-period budgets, portfolio-level thresholds, reserve pools, policy-based release, reallocation approvals, lock windows, and start-prevention controls.","Its reserve mechanism governs cloud spending and runtime capacity rather than reserving AI-pilot capacity for harm investigation."]}],"proximity":"ADJACENT_PRIOR_ART","closest_analogues":[{"name":"UNFPA ring-fenced evaluation funding","similarity":"It places protected evaluation resources within an annual operating ceiling, prohibits diversion to delivery work, and defines formal reallocation authority—the closest structural match to a protected organizational void.","remaining_difference":"It concerns programme evaluation in international development, not a minimum share of AI pilot capacity reserved for predeployment harm inquiry.","source_ids":["C3"]},{"name":"Percentage-based innovation portfolio allocation","similarity":"Innovation-portfolio practice deliberately allocates bounded shares of annual resources to learning-oriented categories and protects uncertain exploration from core delivery pressures.","remaining_difference":"The protected categories seek new products or transformational innovation rather than red-teaming, shadow evaluation, or affected-community inquiry about proposed AI systems.","source_ids":["C4","C5"]},{"name":"Independent AI assurance and three-lines governance","similarity":"AI governance already separates development from independent challenge, requires pre-delivery assessment and approval, and uses audits and red teams to identify risks.","remaining_difference":"These controls are assigned by system or risk tier and do not reserve a stated fraction of organization-year pilot capacity before candidate deployments are selected.","source_ids":["C1","C2","C7"]},{"name":"Anthropic Responsible Scaling Policy","similarity":"The policy institutionalizes recurring evaluations, external review, red teaming, risk reporting, and the possibility of pausing development or constraining deployment.","remaining_difference":"It is capability-threshold governance for frontier models, not a fixed annual portfolio-capacity reserve covering multiple organizational pilots and community inquiry.","source_ids":["C6"]},{"name":"Patented cloud reserve-pool and release governance","similarity":"The patent describes portfolio-level planning-period reserve pools, policy-governed release, approval-controlled reallocation, lock windows, and prevention of new starts.","remaining_difference":"The reserved resource is cloud budget or runtime capacity, and its purpose is cost governance rather than nondeployment investigation of AI harms.","source_ids":["C8"]}],"overlapping_components":["Annual portfolio-level resource allocation","Ring-fenced or protected evaluation resources","Bounded percentage allocation","Independent assurance and effective challenge","Predeployment assessment, red teaming, and approval gates","Learning-oriented experimentation before scaling","Policy-governed release or reallocation criteria","Capacity and time as prerequisites for adequate risk coverage"],"remaining_contrastive_claim":"The remaining distinction is an AI-specific organization-year rule that reserves a stated minimum share of total pilot capacity exclusively for nondeployment harm inquiry before individual pilots are approved, protects that capacity from live-use commitments, and permits release only under explicit criteria.","claim_falsifier":"The contrastive claim would be falsified by a pre-August-2026 AI governance policy, standard, product specification, patent, or documented organizational program that combines a fixed or minimum annual AI-pilot capacity share, exclusive use for red-teaming, shadow evaluation, or affected-community inquiry, protection from deployment work, and explicit release or reallocation rules.","problem_support":"MODERATE","recommendation":"RESEARCH","world_novelty_boundary":"This is a bounded result from eight English-language ordinary-web queries and eight opened sources; it is expressly not a conclusion that the treatment is novel in the world. Proprietary organizational policies, complete patent-family and claims searches, paywalled standards text, non-English sources, and unpublished practices remain outside the search boundary and could contain the same mechanism."}