{ "$schema": "https://json-schema.org/draft/2020-12/schema", "$id": "https://vikramjha.work/press/2026-09-11/artifacts/ai-boardroom-gff-artifact-b.json", "title": "The authority-and-consequence record (The Operator's Map, GFF 2026 Special)", "description": "One record per autonomous action taken by an AI system. Companion artifact to 'The model said so' is not an answer, The Operator's Map, GFF 2026 Special, 11 September 2026, The AI Boardroom (https://vikramjha.work/writing/the-model-said-so-is-not-an-answer). Every property's description names the instrument that asks for it, fetched from its issuer on 11 September 2026. 'Draft' = RBI draft Guidance on Regulatory Principles for Model Risk Management (rbi.org.in Id 5089), a draft on that date. 'Committee' = RBI FREE-AI Committee report, 13 August 2025 (rbi.org.in ID 1306), recommendations, never a rule. '16C' = SEBI (Intermediaries) Regulations, Regulation 16C, in force from 10 February 2025. 'DG' = RBI Deputy Governor's address, 9 September 2026 (Id 1577). 'Governor' = RBI Governor's keynote, 10 September 2026 (Id 1578). Fields whose content is anchored only in an outlet's account carry x-anchor_class 'reported'; all others are 'primary'. No firm, product or client is named.", "type": "object", "required": ["action_id", "system", "authority", "provider", "validation", "oversight", "consequence"], "properties": { "action_id": { "type": "string", "description": "Key. One record per action the system took without a person completing it.", "x-anchor_class": "primary" }, "system": { "type": "object", "required": ["inventory_ref", "acts", "risk_tier"], "properties": { "inventory_ref": { "type": "string", "description": "Reference to the system's row on the model inventory. Draft para 21: 'no model is used, relied upon, or deployed unless it is part of inventory.' Committee Rec 23: an inventory of 'all models, use cases, target groups, dependencies, risks and grievances'.", "x-anchor_class": "primary" }, "acts": { "type": "boolean", "description": "True if the system takes actions (pays, files, sends, closes, approves) rather than only recommending. Draft para 52: consider 'the extent of reliance and the level of autonomy placed on the model outputs for decision-making'. DG para 29 distinguishes a summarizing tool from 'a system that autonomously approves credit or executes financial transactions'.", "x-anchor_class": "primary" }, "risk_tier": { "type": "string", "enum": ["high", "medium", "low", "other"], "description": "The tier assigned under the institution's model risk tiering. Draft para 52 (autonomy as a tiering input); draft para 12(1): the board risk committee reviews validation reports of models with ‘high’ or equivalent risk and approves their deployment.", "x-anchor_class": "primary" } } }, "authority": { "type": "object", "required": ["approver", "approval_date", "rationale_ref", "permitted_scope"], "properties": { "approver": { "type": "string", "description": "The person or function that granted approval for deployment. Draft definition (4): 'Model Approver means individual or function, responsible for undertaking approval process and granting approval for model deployment.'", "x-anchor_class": "primary" }, "approval_date": { "type": "string", "format": "date", "description": "Date of the deployment approval. Draft para 35: 'The decision-making process and exception approval for a model should be documented and should include the rationale for decision / approval.'", "x-anchor_class": "primary" }, "rationale_ref": { "type": "string", "description": "Reference to the written rationale for the approval. Draft para 35 ('the rationale for decision / approval').", "x-anchor_class": "primary" }, "exception": { "type": "boolean", "description": "True if the deployment was approved as an exception to the framework. Draft para 35 ('exception approval for a model should be documented'); draft para 12(3): the board risk committee oversees 'models approved with exceptions'.", "x-anchor_class": "primary" }, "permitted_scope": { "type": "object", "required": ["limit", "enforced_where"], "properties": { "limit": { "type": "string", "description": "The largest action the system may take unaided, in the unit that applies (amount, count, class of action)." }, "enforced_where": { "type": "string", "enum": ["rail", "model", "orchestration"], "description": "Where the limit is enforced. Draft para 60(ii); DG para 29. A limit enforced at the rail is a control; a limit the model reasons about is a suggestion (the Ship AI chapter of this special)." } }, "description": "The permission and limit record. Draft para 60(ii): 'override, suspension, or deactivation mechanisms, including kill-switch arrangements'; DG para 29 (proportionality by consequence).", "x-anchor_class": "primary" } } }, "owner": { "type": "string", "description": "The person or function responsible for the model across its lifecycle. Draft definition (6): 'Model Owner means an individual or function, responsible for ensuring that the model's design, assumptions, methodologies, and documentation, are aligned with its intended use'.", "x-anchor_class": "primary" }, "provider": { "type": "object", "required": ["developer_or_provider", "third_party"], "properties": { "developer_or_provider": { "type": "string", "description": "Who designed, developed, tested and trained the model. Draft definition (5) 'Model Developer'; draft para 8: accountable 'irrespective of whether the models are developed internally, sourced from third-parties, or a combination thereof'; 16C(1): 'either designed by it or procured from third-party technology service providers'.", "x-anchor_class": "primary" }, "third_party": { "type": "boolean", "description": "True if any material component is sourced from a third party. Draft para 8; 16C(1). This is the column for the risk the Governor's para 17 names as 'concentration and herding' and DG para 20 describes as a common dependency across institutions.", "x-anchor_class": "primary" }, "auto_update_scope": { "type": "string", "description": "What the provider may change without the institution's approval. Draft para 56: 'defining a clear scope of what can be updated automatically, strict justifications for enabling automatic updates'.", "x-anchor_class": "primary" }, "undisclosed_items": { "type": "array", "items": { "type": "string" }, "description": "What the provider did not disclose that limits the institution's own validation. Draft para 53: 'limitations in independent validation'.", "x-anchor_class": "primary" }, "last_provider_update": { "type": "string", "format": "date", "description": "Date of the most recent provider-driven change. Draft para 53: “changes in model behaviour or capabilities resulting from provider-driven updates”. Committee 4.4.68 'Dependencies:' names 'Third-party providers, cloud service providers, data sources, and any other external components that can influence AI model performance.'", "x-anchor_class": "primary" } } }, "validation": { "type": "object", "required": ["validator", "report_date"], "properties": { "validator": { "type": "string", "description": "The independent validator. Draft definition (8): 'Model Validator means an individual or function, independent of model development or ownership or use'; draft para 29: all models, 'including third-party models, are subject to independent validation by the RE'.", "x-anchor_class": "primary" }, "report_date": { "type": "string", "format": "date", "description": "Date the validation was completed. Draft para 33.", "x-anchor_class": "primary" }, "rmcb_date": { "type": "string", "format": "date", "description": "Date the validation report was placed before the board's risk committee. Draft para 33: 'within three months of completion of the validation'.", "x-anchor_class": "primary" }, "explainability_threshold": { "type": "string", "description": "The explainability threshold set for this system. Draft para 54(1)(i): 'apply higher thresholds for explainability to models which are relied upon for material decision-making or have significant impact on customers or its operations'. Governor para 17 names 'opacity' first among the risks.", "x-anchor_class": "primary" }, "explainability_met": { "type": "boolean", "description": "Whether the output of this action met the threshold. Draft para 54(1)(i).", "x-anchor_class": "primary" } } }, "oversight": { "type": "object", "required": ["mode"], "properties": { "mode": { "type": "string", "enum": ["human_in_loop", "human_on_loop", "other"], "description": "The human-in-command arrangement. Draft para 60(i): 'Human-in-command arrangements (e.g., human-in-the-loop / human-on-the-loop / other human oversight mechanisms)'. This and the override field are the columns for the risk the Governor's para 17 names as 'erosion of human judgement'. No field in this schema covers bias, cybersecurity or data privacy; the chapter says so.", "x-anchor_class": "primary" }, "override": { "type": "object", "properties": { "occurred": { "type": "boolean" }, "by": { "type": "string" }, "at": { "type": "string", "format": "date-time" }, "reason": { "type": "string" } }, "description": "Any human override of this action. Draft para 60(ii); draft para 63 ('decisions, interventions, overrides, incidents and near misses'); Committee Sutra 2: humans 'should be able to override AI'.", "x-anchor_class": "primary" }, "kill_switch_tested_at": { "type": "string", "format": "date", "description": "When the deactivation mechanism was last exercised. Draft para 60(ii) ('kill-switch arrangements'); Committee 4.4.74 ('mechanisms exist to stop, pause or unwind AI-driven processes in a controlled manner').", "x-anchor_class": "primary" }, "kill_switch_measured_seconds": { "type": "number", "description": "Measured time from the switch to the last dependent system stopping. Draft para 60(ii).", "x-anchor_class": "primary" } } }, "consequence": { "type": "object", "required": ["outcome", "customer_affected"], "properties": { "outcome": { "type": "string", "description": "What the action did. Draft para 8: 'An RE is accountable for the outcomes of all models used by it'; DG para 21: 'it cannot outsource the consequence'.", "x-anchor_class": "primary" }, "customer_affected": { "type": "boolean", "description": "Whether a customer was affected. DG para 21: 'A customer affected by an important financial decision deserves something more meaningful than being told that the model said so.'", "x-anchor_class": "primary" }, "incident_or_near_miss": { "type": "string", "enum": ["none", "near_miss", "incident"], "description": "Draft para 63 ('incidents and near misses'); Committee 4.4.48: “Any errant model behaviour or incidents must be formally recorded and reported through appropriate channels”; Committee Rec 22 (a sectoral AI incident reporting framework, recommended to regulators).", "x-anchor_class": "primary" }, "reversible": { "type": "boolean", "description": "Whether the action could be undone. Committee 4.4.74: the audit 'should also confirm that mechanisms exist to stop, pause or unwind AI-driven processes in a controlled manner'. As reported by MediaNama, the SEBI Chairman named 'reversibility' on the festival panel; that is the outlet's account, beside the primary anchor.", "x-anchor_class": "primary" }, "reversed_at": { "type": "string", "format": "date-time", "description": "When it was reversed, if it was. Committee 4.4.74.", "x-anchor_class": "primary" }, "customer_disclosed_ai": { "type": "boolean", "description": "Whether the customer was told they were interacting with an AI system. Draft para 59(ii): 'appropriate disclosures and warnings should be provided to the users that they are interacting with AI / ML based system'.", "x-anchor_class": "primary" }, "customer_human_option": { "type": "boolean", "description": "Whether the customer could switch to a person. Draft para 59(iii): 'option should be provided to the customer to switch to human assistance when requested for.'", "x-anchor_class": "primary" }, "reproducibility_ref": { "type": "string", "description": "Reference that allows the action to be reproduced from its inputs. Draft para 57: documentation 'to enable traceability, reproducibility, and auditability'.", "x-anchor_class": "primary" }, "grievances_ref": { "type": "string", "description": "Reference to any grievance filed in respect of this system or action. Committee 4.4.68 'Grievances:': 'A record of the volume and nature of grievances filed in respect of these AI systems and how they were resolved.'", "x-anchor_class": "primary" } } }, "agent": { "type": "object", "description": "The acting agent's own identity, authentication and consent scope. The primary anchors name only the SCOPE risk: Committee 4.4.63 ('AI agents may act beyond their intended scope;') and 4.4.74 (audit coverage of 'agent-to-agent interactions'). The CONTENT of these three fields (agent identity, authentication, customer consent) is described only in an outlet's account of a bank chairman's festival remarks (Business Standard, 10 September 2026: mechanisms 'covering agent identity, authentication, customer consent, transaction limits, audit trails and location'); no record of those remarks was found on the bank's site as of 11 September 2026. The Ship AI chapter of this special carries the identity argument.", "properties": { "identity": { "type": "string", "description": "As reported (Business Standard, IANS, 10 September 2026). Primary anchor for scope only: Committee 4.4.63, 4.4.74.", "x-anchor_class": "reported" }, "authenticated_by": { "type": "string", "description": "As reported (Business Standard, IANS, 10 September 2026). Primary anchor for scope only: Committee 4.4.63, 4.4.74.", "x-anchor_class": "reported" }, "consent_ref": { "type": "string", "description": "As reported (Business Standard, IANS, 10 September 2026). Primary anchor for scope only: Committee 4.4.63, 4.4.74.", "x-anchor_class": "reported" } } }, "liability": { "type": "object", "properties": { "responsible_entity": { "type": "string", "description": "The regulated entity that carries the consequence. 16C(1): 'shall be solely responsible' for 'the output arising from the usage of such tools and techniques it relies upon or deals with'. Draft para 8. Committee Sutra 5: 'Accountability cannot be delegated to the model and underlying algorithm.'", "x-anchor_class": "primary" }, "examined_by": { "type": "string", "enum": ["RBI", "SEBI", "IRDAI", "other"], "description": "The supervisor whose examination this record must survive. Draft paras 8-9 (an RE under the RBI's guidance); 16C ('Any person regulated by the Board'); IRDAI office order of 17 June 2026 (regulated entities in the insurance sector).", "x-anchor_class": "primary" } } } }, "x-status": { "as_of": "2026-09-11", "note": "The RBI draft is a draft on this date (header 'June xx, 2026'; 'the draft framework', Governor, 10 September 2026; listed on the RBI's Draft Notifications / Guidelines page on 11 September 2026). Paragraph numbers must be re-checked against the final text when it issues. The FREE-AI report is a set of recommendations. SR 11-7 is superseded (17 April 2026) and is not an anchor here. Nothing in this schema names a firm, product, client or incident." } }