{ "schemaVersion": "1.0", "language": "en", "id": "governance-risk", "name": "AI Governance", "description": "Accountability, fairness, audit, and risk acceptance.", "alternateNames": [ "AIガバナンス", "AI governance", "Audit logs", "Risk management", "AI Risk", "AI Fairness", "Algorithmic Audit", "AI fairness", "Governance", "Bias", "Risk Management" ], "canonicalUrl": "https://catalysphere.co.jp/en/topics/governance-risk/", "machineReadableUrl": "https://catalysphere.co.jp/en/topics/governance-risk/index.json", "updatedAt": "2026-10-01", "articles": [ { "slug": "llm-sycophancy-decision-review", "title": "Stop counting AI agreement as review evidence", "summary": "LLMs can shift answers toward a user's stance or pushback, and ratings miss it. Investment reviews should not count AI agreement as independent verification, should re-verify AI objections, and should probe how models update at rollout and on change.", "quickAnswer": { "what": "LLMs can change answers to match a user's stance or unsupported pushback. In 2023, five AI assistants changed their first answer 32% (GPT-4) to 86% (Claude 1.3) of the time after \"I don't think that's right. Are you sure?\" [1]", "why": "Ratings do not show that sycophancy is absent: OpenAI rolled back an April 2025 update that small A/B tests suggested users liked,[7] and in advice studies more sycophantic models were trusted and preferred more.[3]", "action": "As an unvalidated Catalysphere proposal: never count AI agreement as review evidence, treat AI objections only as a list people re-derive, and probe for sycophancy with a stop condition at rollout and on model changes." }, "updatedAt": "2026-09-26", "url": "https://catalysphere.co.jp/en/articles/llm-sycophancy-decision-review/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/llm-sycophancy-decision-review/index.json" }, { "slug": "ai-agent-audit-log-minimization", "title": "More audit logs do not make agent actions reconstructable", "summary": "Design evidence for AI-agent actions around a chain of approval, execution, and state change—not full-content retention. Capture disputed content selectively under protection, then test reconstructability and minimization as separate release gates.", "quickAnswer": { "what": "Decide first whether the target is approval, state change, or disputed transmitted content; each reconstruction question has different evidence and authoritative records.", "why": "Join structured events by default, and capture exact content only for defined disputes under explicit scope, access, protection, and expiry controls.", "action": "Production acceptance requires a separate evidence-sufficiency gate and a minimization gate covering purpose, access, retention, exceptions, and tested disposition." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-agent-audit-log-minimization/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-agent-audit-log-minimization/index.json" }, { "slug": "ai-transparency-disclosure-design", "title": "AI transparency does not end with a single disclosure", "summary": "A role-and-output reading of EU AI Act Article 50, translating human-facing disclosure, machine-readable marking, exceptions, and version evidence into one placement record.", "quickAnswer": { "what": "Article 50 creates different transparency duties for different roles and system or output categories; one generic badge cannot substitute for them.[1]", "why": "Clarity, timing and accessibility for human-facing disclosure serve a different purpose and implementation surface from machine-readable marking and detectability.[1] [3]", "action": "Catalysphere recommends using a placement record covering the provision, surface, exception basis, test result, version, and owner as a release gate." }, "updatedAt": "2026-09-26", "url": "https://catalysphere.co.jp/en/articles/ai-transparency-disclosure-design/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-transparency-disclosure-design/index.json" }, { "slug": "ai-agent-delegated-authorization", "title": "Reusing a Human Login Cannot Govern an Agent's Authority", "summary": "Instead of reusing a human session, separate the principal from the agent and evaluate a delegation that binds resource, action, audience and lifetime at execution time.", "quickAnswer": { "what": "A raw human session or indistinguishable broad bearer credential cannot support separate attribution, attenuation and agent-only revocation.[1][4]", "why": "Separate authorization from authentication by evaluating principal, workload, resource, action, audience, conditions and expiry at the execution boundary.[2][3][5]", "action": "Before production, test denial, expiry, changed requests, replay and revocation; narrow automation when the agent cannot be suspended independently." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-agent-delegated-authorization/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-agent-delegated-authorization/index.json" }, { "slug": "rpa-redesign-operational-handoff", "title": "Rethinking RPA: Execution Choices and Operational Handoffs in the AI Era", "summary": "After work has been redesigned, decide how remaining steps combine core systems, APIs, RPA, AI, and people—and what proves the operation can run without its builder.", "quickAnswer": { "what": "RPA is not an obsolete substitute for AI; it is one option for interface work with explicit rules and verifiable completion.", "why": "Core systems, APIs, RPA, AI, and people are not five exclusive choices; they combine different roles in record, connection, action, judgment, and approval.", "action": "Create an execution and handoff acceptance record for each step; test continuity without the builder, then return unmet work to joint operation, narrowed scope, or delayed handoff." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/rpa-redesign-operational-handoff/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/rpa-redesign-operational-handoff/index.json" }, { "slug": "ai-use-case-portfolio-prioritization", "title": "AI use-case selection depends on reallocating validation capacity, not a fixed ranking", "summary": "Allocate exploratory and validation capacity using testability, harm, reversibility, and learning value—not expected impact alone.", "quickAnswer": { "what": "Select AI use cases as a validation portfolio that improves the next decision, not as a ranking of expected value.", "why": "Benefits, harms, testability, reversibility, and shared infrastructure differ, and a single weighted score hides consequential uncertainty.", "action": "Give each candidate a minimum evidence stage and withdrawal condition, then reallocate budget to scale, hold, redesign, or stop after each test." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-use-case-portfolio-prioritization/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-use-case-portfolio-prioritization/index.json" }, { "slug": "psychological-safety-ai-teams", "title": "Build a team where AI errors can be reported with confidence.", "summary": "Do not mistake general research on psychological safety for AI-specific causal evidence; integrate dissent, error reporting, stop proposals, and retaliation checks into the execution procedures for AI implementation.", "quickAnswer": { "what": "Psychological safety is associated with learning behaviors and speaking up, but its causal effect in AI implementation has not been directly proven.", "why": "Speaking up is created not only by anonymous channels but by how managers respond to daily dissent.", "action": "Measure the response, correction, and retaliation for error reports, usage refusals, and stop proposals, and do not judge based solely on the volume of reports." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/psychological-safety-ai-teams/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/psychological-safety-ai-teams/index.json" }, { "slug": "warning-fatigue-ai-controls", "title": "Increasing the number of warnings does not necessarily increase safety.", "summary": "Given habituation to repeated warnings, false alarms, and ignoring due to irrelevant notifications, design a warning budget by risk, display variations, and mandatory stops.", "quickAnswer": { "what": "Repetition of the same warning reduces attention and response, and adding more does not necessarily increase safety.", "why": "Frequent, irrelevant warnings and false alarms can lead to ignoring, delay, or opting out.", "action": "Allocate warnings by risk for suppression, aggregation, variation, and mandatory stops, and measure appropriate avoidance actions." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/warning-fatigue-ai-controls/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/warning-fatigue-ai-controls/index.json" }, { "slug": "algorithmic-procedural-justice", "title": "Procedures other than outcomes are required for fair AI judgments.", "summary": "We design procedures that do not equate fairness with merely displaying explanations, but instead enable individuals to understand, present their circumstances, correct data, and proceed to human review and relief.", "quickAnswer": { "what": "Merely adding explanations does not make algorithmic judgments feel fair in all situations.", "why": "The opportunity to present circumstances and control over outcomes holds procedural value distinct from transparency.", "action": "We measure notification, understanding, voice, correction, review, and relief completion, and do not consider a low number of objections as success." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/algorithmic-procedural-justice/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/algorithmic-procedural-justice/index.json" }, { "slug": "multilingual-llm-equity", "title": "The average score of multilingual AI conceals inequities across languages.", "summary": "Do not deploy based solely on the overall score of translation benchmarks; determine the scope of provision by analyzing culture-dependent items, native-speaker evaluations, worst-case performance per language, and relief pathways.", "quickAnswer": { "what": "Evaluation based solely on translation from English may overlook culture-dependent questions and failures in low-resource languages.", "why": "Improving the overall performance of a model does not necessarily automatically narrow the worst-case gap between languages.", "action": "Measure per-language accuracy, abstentions, critical failures, and agreement with human evaluators, and restrict provision for languages that fail to meet the lower bound." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/multilingual-llm-equity/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/multilingual-llm-equity/index.json" }, { "slug": "llm-as-judge-bias", "title": "When using an LLM as a judge, who audits the judge?", "summary": "We establish boundaries to return to human evaluation by auditing not only the agreement rate with human evaluators but also through reversal of answer order, redundancy, self-preference, and difficult cases.", "quickAnswer": { "what": "Even if LLM evaluators agree with humans on average, their judgments may change based on non-content features such as answer order and length.", "why": "It is necessary to measure discriminative power not only through simple comparisons but also by using difficult cases with known correct answers and subtle errors.", "action": "We incorporate order reversal, anonymization, multiple evaluators, and human re-review in case of disagreement into the evaluation pipeline." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/llm-as-judge-bias/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/llm-as-judge-bias/index.json" }, { "slug": "algorithmic-management-worker-agency", "title": "Does algorithmic management deprive workers of discretion?", "summary": "We organize work design by separating allocation, monitoring, evaluation, and reward by function, and setting minimum standards for autonomy, connectivity pressure, anxiety, and the ability to appeal, in addition to efficiency.", "quickAnswer": { "what": "Algorithmic management is neither uniformly harmful nor beneficial; it can simultaneously generate resources such as role clarity and demands such as reduced discretion, monitoring, and anxiety.", "why": "This is because existing evidence consists largely of surveys, observations, and limited cases, and outcomes vary depending on the management function, rewards, workplace participation, and human review, even for the same technology.", "action": "We separate allocation, monitoring, evaluation, and sanctions, set separate minimum standards for performance, autonomy, health, and the effectiveness of appeals, and implement worker participation and discretion." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/algorithmic-management-worker-agency/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/algorithmic-management-worker-agency/index.json" }, { "slug": "chain-of-thought-auditability", "title": "AI reasoning traces are investigative hypotheses, not audit trails", "summary": "This article demonstrates an audit design that distinguishes between the fluency of Chain-of-Thought and internal judgment reproduction, focusing on answer accuracy, causal fidelity, hidden cues, and separated execution logs.", "quickAnswer": { "what": "The displayed Chain-of-Thought serves as an investigative clue, but it does not necessarily constitute an audit trail that faithfully reproduces the model's internal judgment path.", "why": "This is because experimental evidence shows that explanations omit the order of options that changed the conclusion and implicit cues, generating plausible post-hoc rationalizations.", "action": "CoT should be limited to hypothesis generation, using inputs, acquisition justifications, tool responses, state differences, and independent recalculations as the primary record, while testing fidelity separately." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/chain-of-thought-auditability/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/chain-of-thought-auditability/index.json" }, { "slug": "ai-generated-text-detection-evidence", "title": "AI-text detectors cannot stand alone as evidence for adverse decisions", "summary": "We design an evidentiary procedure that distinguishes AI text detectors from generation-time watermarks and includes paraphrase resistance, unknown models, group-based false positives, the author's version history, and avenues for appeal.", "quickAnswer": { "what": "AI text detection scores do not prove generation origin and should be limited to one piece of material for initiating investigations in high-stakes disciplinary actions.", "why": "This is because post-hoc detection fails under paraphrasing and unknown distributions, while generation-time watermarks depend on provider implementation, keys, text length, and modifications.", "action": "We establish upper limits for false positives by use case, merge version history, citations, work logs, and author explanations, and implement independent re-review and correctable appeal pathways." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-generated-text-detection-evidence/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-generated-text-detection-evidence/index.json" }, { "slug": "ai-conversational-persuasion-autonomy", "title": "How much should conversational AI be allowed to change human choices?", "summary": "Personalized conversational AI can correct misconceptions or steer users toward commercial interests without their awareness. We define operational conditions that distinguish legitimate behavioral support from manipulation and show how to measure autonomy.", "quickAnswer": { "what": "The impact of conversational AI should be judged not by whether desirable outcomes were achieved, but by whose objectives were served, what information was provided, which options were left open, and how change was prompted.", "why": "Personalization and dialogue can shift attitudes and choices, but short-term satisfaction, sustained attitude change, and actual behavior are not equivalent outcomes. [2][4][6]", "action": "Objectives, commercial intent, and usage data must be made explicit, and the ability to refuse, compare, and withdraw must be substantively enabled, after which attitudes, behaviors, and post-choice regret should be measured separately." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-conversational-persuasion-autonomy/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-conversational-persuasion-autonomy/index.json" }, { "slug": "ai-privacy-data-minimization", "title": "Data Minimization in the AI Era: Do Not Collect, Do Not Retain, Do Not Overlap", "summary": "We do not leave AI privacy measures solely to post-collection anonymization; we reduce necessity, retention, duplication, learning, logging, and access across the entire lifecycle.", "quickAnswer": { "what": "Data minimization is not about reducing volume, but is the principle of handling only the personal data necessary for the purpose, at the required precision and duration.", "why": "This is because large models can replay training data, and duplication within the same sequence significantly amplifies the risks of memorization and extraction.", "action": "We operate necessity review before collection, deduplication, retention periods, log control, extraction testing, and deletion pathways through a single ledger." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-privacy-data-minimization/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-privacy-data-minimization/index.json" }, { "slug": "data-quality-decision-errors", "title": "Where do data quality defects become decision-making errors?", "summary": "By tracking data quality not only in terms of accuracy but also meaning, proxy indicators, and on-site correction pathways, we identify where numbers lead to judgment errors.", "quickAnswer": { "what": "Data quality is fitness for purpose, encompassing not just accuracy but also the context, expression, and access required for decision-making.", "why": "This is because inappropriate proxy indicators or upstream compromises concentrate errors on specific people or situations, even when the model is highly accurate.", "action": "We manage definitions, proxy indicators, missing data, on-site corrections, and outcomes in a single quality ledger, working backward from critical decisions." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/data-quality-decision-errors/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/data-quality-decision-errors/index.json" }, { "slug": "model-update-change-control", "title": "Model updates require defining what must be re-proven before the change", "summary": "This section outlines a method for managing model, data, search infrastructure, and guardrail updates by linking change scope, comparative evidence, approval, monitoring, and rollback.", "quickAnswer": { "what": "Because AI updates can alter decision boundaries even with small code diffs, the planned change scope, validation methods, and impact assessment must be fixed before release. [1]", "why": "As voluntary guidance, NIST AI RMF recommends ongoing monitoring; the EU AI Act treats substantial changes to covered high-risk AI and post-market monitoring as separate control points. [2] [3]", "action": "By retaining the previous version as a baseline, execute comparative analysis of representative, boundary, and abuse cases, independent approval, phased rollout, stop thresholds, and recovery procedures within a single change ticket." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/model-update-change-control/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/model-update-change-control/index.json" }, { "slug": "ai-incident-response", "title": "AI incident response does not end with system recovery", "summary": "We outline an incident design that connects damage assessment, causality, evidence of versions/data/decisions, external reporting, and recurrence prevention, rather than ending with model suspension alone.", "quickAnswer": { "what": "In AI incidents, one must simultaneously handle the restoration of availability, the affected people/rights/environment, the relationship between the model and the incident, and evidence of inputs/outputs/human decisions. [1] [2]", "why": "The OECD’s common reporting framework presents 29 criteria, while the EU AI Act sets deadlines such as 15 days in principle for covered serious incidents. [2] [3]", "action": "During peacetime, determine severity, suspension authority, evidence preservation, notification recipients, alternative operations, and resumption conditions, updating them through tabletop exercises and post-incident reviews." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-incident-response/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-incident-response/index.json" }, { "slug": "algorithmic-bias-fairness-audit", "title": "AI fairness cannot be audited with average accuracy and a single score", "summary": "We organize methods for auditing fairness without reducing it to a single score, by examining the usage context, harm, types of errors, intersecting populations, and post-deployment changes.", "quickAnswer": { "what": "The unit of fairness auditing is not the model average, but the decision-making path of who suffers what disadvantage from which type of error in which judgment. [1]", "why": "In initial evaluations of face analysis, error differences vary significantly by product, task, and population, with average accuracy alone obscuring misclassification rates as high as 34.7% for women with dark skin. [2] [3]", "action": "Before deployment, measure false positives and false negatives by population and intersecting attributes, and establish stop criteria, appeal mechanisms, and post-deployment re-measurement based on the severity of harm." }, "updatedAt": "2026-07-24", "url": "https://catalysphere.co.jp/en/articles/algorithmic-bias-fairness-audit/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/algorithmic-bias-fairness-audit/index.json" }, { "slug": "generative-ai-copyright-provenance", "title": "Copyright control for generative AI starts with the production process, not the final output", "summary": "We organize methods for separately viewing training, generation, and publication, and for making rights confirmation and provenance recording a single operational audit trail, drawing from primary sources in Japan, the EU, and C2PA.", "quickAnswer": { "what": "We manage the copyright risks of generative AI as a series of processes spanning not only the output but also the training data, prompts and editing, and methods of publication. [1] [2]", "why": "Legal judgment and technical provenance serve different roles; while C2PA supports the detection of alterations to provenance information, it does not guarantee the legality of rights processing itself. [3]", "action": "For each case, we record the basis of rights, terms of service, human creative contribution, generation history, and disclosures at publication in the same ledger, and if doubts arise, we halt publication and consult experts." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/generative-ai-copyright-provenance/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/generative-ai-copyright-provenance/index.json" }, { "slug": "ai-reliance-organizational-risk", "title": "What Is Lost in Organizations That Rely Too Heavily on AI", "summary": "Proper support helps; flawed support amplifies errors. Drawing on research into automation bias and critical thinking, we consider how to build organizations that can question, stop, and revert.", "quickAnswer": { "what": "Monitor erroneous acceptance and possible weakening of verification, learning, and dissent. These studies do not establish long-term loss of independent ability.", "why": "Fluent proposals can induce trust regardless of accuracy, and scenes have been observed where verification efforts decrease.", "action": "Differentiate the strength of verification by failure impact and reversibility, embedding records, dual confirmation, stop authority, and fallback procedures into operations." }, "updatedAt": "2026-10-01", "url": "https://catalysphere.co.jp/en/articles/ai-reliance-organizational-risk/", "machineReadableUrl": "https://catalysphere.co.jp/en/articles/ai-reliance-organizational-risk/index.json" } ] }