{
  "object_name": "Agent Accountability Checklist",
  "slug": "agent-accountability-checklist",
  "object_type": "LPM Knowledge Object",
  "version": "1.0",
  "last_reviewed": "2026-06-24",
  "owner": "LPM / Lapemo",
  "primary_lpm_layer": "AI Amplification",
  "connected_layers": [
    "Ownership Map",
    "Decision Architecture",
    "Information Ecology",
    "Platform Structure",
    "Governance Architecture",
    "Control Map",
    "AI Governance Checklist"
  ],
  "purpose": "Use this Agent Accountability Checklist before any AI agent is piloted, connected to tools, allowed to retrieve enterprise knowledge, allowed to communicate, or allowed to act across systems. The checklist makes one thing explicit: an agent cannot be accountable. People are accountable for the agent, its scope, its data, its tools, its actions, its evidence, its controls, and its outcomes. This object helps companies scale AI without creating hidden automation, invisible decision rights, uncontrolled system actions, or audit gaps.",
  "intended_audience": [
    "Executives",
    "Transformation leaders",
    "AI governance leaders",
    "Product leaders",
    "Platform owners",
    "Risk and control owners",
    "Data leaders",
    "Technology leaders",
    "Agent builders"
  ],
  "agent_types": [
    {
      "type": "Assistant agent",
      "description": "Answers questions, drafts content, summarizes information, or supports a human user without system action."
    },
    {
      "type": "Workflow agent",
      "description": "Coordinates tasks, routes work, prepares handoffs, checks status, or moves work across teams and systems."
    },
    {
      "type": "Decision-support agent",
      "description": "Recommends, scores, prioritizes, flags, classifies, or prepares options for a human decision owner."
    },
    {
      "type": "System-action agent",
      "description": "Updates records, triggers workflows, sends messages, opens tickets, modifies data, or invokes APIs."
    },
    {
      "type": "Customer or employee-facing agent",
      "description": "Interacts directly with customers, employees, vendors, candidates, or partners."
    },
    {
      "type": "Multi-agent chain",
      "description": "Uses multiple agents, models, tools, or handoffs where accountability can become fragmented."
    }
  ],
  "required_fields": [
    {
      "field": "Agent ID",
      "definition": "Unique identifier for the agent, agent workflow, or agent chain",
      "required": "Yes"
    },
    {
      "field": "Agent name",
      "definition": "Plain-language name of the agent",
      "required": "Yes"
    },
    {
      "field": "Agent type",
      "definition": "Assistant, workflow, decision-support, system-action, customer-facing, employee-facing, or multi-agent chain",
      "required": "Yes"
    },
    {
      "field": "Business purpose",
      "definition": "Outcome, workflow, service, decision, control, or operating need the agent supports",
      "required": "Yes"
    },
    {
      "field": "Accountable business owner",
      "definition": "Person or role accountable for agent outcome, use, risk, value, and lifecycle",
      "required": "Yes"
    },
    {
      "field": "Technical owner",
      "definition": "Person or role accountable for implementation, reliability, logs, integrations, and operational health",
      "required": "Yes"
    },
    {
      "field": "Data or knowledge owner",
      "definition": "Person or role accountable for approved sources, retrieval quality, sensitivity, access, freshness, and lineage",
      "required": "Yes"
    },
    {
      "field": "Control owner",
      "definition": "Person or role accountable for controls, evidence, monitoring, exceptions, and testing",
      "required": "Yes"
    },
    {
      "field": "Decision owner",
      "definition": "Person or forum accountable for decisions influenced by the agent",
      "required": "Required when decision-support exists"
    },
    {
      "field": "Approved tasks",
      "definition": "Tasks the agent is allowed to perform",
      "required": "Yes"
    },
    {
      "field": "Blocked tasks",
      "definition": "Tasks the agent is not allowed to perform",
      "required": "Yes"
    },
    {
      "field": "Approved tools and systems",
      "definition": "Applications, APIs, integrations, workflows, repositories, and channels the agent may use",
      "required": "Yes"
    },
    {
      "field": "Blocked tools and systems",
      "definition": "Applications, APIs, integrations, channels, or data stores the agent may not use",
      "required": "Yes"
    },
    {
      "field": "Allowed actions",
      "definition": "Retrieve, summarize, draft, recommend, score, route, update, trigger, communicate, execute, or escalate",
      "required": "Yes"
    },
    {
      "field": "Blocked actions",
      "definition": "Approvals, commitments, external communication, data changes, control bypass, policy exceptions, regulated decisions, or irreversible actions without approval",
      "required": "Yes"
    },
    {
      "field": "Human review rule",
      "definition": "Who reviews, when review is required, what evidence is reviewed, and what authority the reviewer has",
      "required": "Yes"
    },
    {
      "field": "Evidence and logging rule",
      "definition": "What prompts, context, sources, outputs, tool calls, approvals, exceptions, and actions must be logged",
      "required": "Yes"
    },
    {
      "field": "Impact tier",
      "definition": "Low, moderate, high, critical, regulated, customer-facing, employee-impacting, financial, security, privacy, or control-impacting",
      "required": "Yes"
    },
    {
      "field": "Monitoring signals",
      "definition": "Accuracy, drift, misuse, incidents, overrides, adoption, value, control failures, latency, access exceptions, and user feedback",
      "required": "Yes"
    },
    {
      "field": "Escalation path",
      "definition": "Where issues go when agent output, action, access, or behavior becomes unsafe, stale, incorrect, unauthorized, or outside scope",
      "required": "Yes"
    },
    {
      "field": "Kill switch or pause owner",
      "definition": "Person or role authorized to pause, restrict, roll back, or retire the agent",
      "required": "Required for moderate and above"
    },
    {
      "field": "Lifecycle status",
      "definition": "Idea, design, pilot, limited release, production, scaled, restricted, paused, retired, or superseded",
      "required": "Yes"
    },
    {
      "field": "Review cadence",
      "definition": "Weekly, monthly, quarterly, release-based, incident-based, policy-change based, source-change based, model-change based, or access-change based",
      "required": "Yes"
    }
  ],
  "checklist_sections": [
    {
      "section": "1. Agent identity and purpose",
      "items": [
        {
          "item": "Agent has a unique ID and clear name",
          "status": "Yes / No / Partial",
          "guidance": "Avoid unnamed automations, prompt collections, or embedded vendor features with no owner."
        },
        {
          "item": "Agent type is classified",
          "status": "Yes / No / Partial",
          "guidance": "Classify the strongest role the agent can perform, not the softest description."
        },
        {
          "item": "Business purpose is tied to a measurable outcome",
          "status": "Yes / No / Partial",
          "guidance": "Tie the agent to value, risk reduction, cycle time, quality, control strength, or customer/employee outcome."
        },
        {
          "item": "Agent is linked to a workflow, decision, service, product, control, or operating-model object",
          "status": "Yes / No / Partial",
          "guidance": "Agents should not float outside the operating model."
        }
      ]
    },
    {
      "section": "2. Human accountability model",
      "items": [
        {
          "item": "One accountable business owner is named",
          "status": "Yes / No / Partial",
          "guidance": "No shared-accountability fog. One person or role owns the outcome."
        },
        {
          "item": "Technical owner is named",
          "status": "Yes / No / Partial",
          "guidance": "Technical ownership covers reliability, integrations, logs, incidents, and change management."
        },
        {
          "item": "Data or knowledge owner is named",
          "status": "Yes / No / Partial",
          "guidance": "Retrieval quality and source freshness need a real owner."
        },
        {
          "item": "Control owner is named",
          "status": "Yes / No / Partial",
          "guidance": "Control ownership covers monitoring, evidence, exceptions, and testing."
        },
        {
          "item": "Decision owner is named when the agent influences decisions",
          "status": "Yes / No / Partial",
          "guidance": "AI cannot become the hidden decision owner."
        }
      ]
    },
    {
      "section": "3. Scope, tasks, and operating boundary",
      "items": [
        {
          "item": "Approved tasks are explicitly defined",
          "status": "Yes / No / Partial",
          "guidance": "List what the agent is allowed to do in plain language."
        },
        {
          "item": "Blocked tasks are explicitly defined",
          "status": "Yes / No / Partial",
          "guidance": "List what the agent must never do without approval."
        },
        {
          "item": "Start and stop conditions are documented",
          "status": "Yes / No / Partial",
          "guidance": "Define when the agent starts work, stops work, hands off, or escalates."
        },
        {
          "item": "Agent handoffs are mapped",
          "status": "Yes / No / Partial",
          "guidance": "Map handoffs to humans, queues, systems, other agents, and escalation paths."
        }
      ]
    },
    {
      "section": "4. Decision and action rights",
      "items": [
        {
          "item": "Allowed actions are classified",
          "status": "Yes / No / Partial",
          "guidance": "Retrieve, summarize, draft, recommend, score, route, update, trigger, communicate, execute, or escalate."
        },
        {
          "item": "Blocked actions are classified",
          "status": "Yes / No / Partial",
          "guidance": "Block approvals, commitments, external messages, regulated decisions, system-of-record changes, and control bypass unless approved."
        },
        {
          "item": "Human approval is required for high-impact actions",
          "status": "Yes / No / Partial",
          "guidance": "Do not rely on vague human-in-the-loop language."
        },
        {
          "item": "Agent cannot expand its own authority",
          "status": "Yes / No / Partial",
          "guidance": "The agent should not grant itself tool access, choose new sources, bypass review, or change its own scope."
        }
      ]
    },
    {
      "section": "5. Data, knowledge, and retrieval boundaries",
      "items": [
        {
          "item": "Approved sources are listed and owned",
          "status": "Yes / No / Partial",
          "guidance": "Map documents, repositories, systems, dashboards, APIs, knowledge objects, and data products."
        },
        {
          "item": "Blocked sources and data classes are listed",
          "status": "Yes / No / Partial",
          "guidance": "Include sensitive, stale, private, unapproved, draft, regulated, or unsupported sources."
        },
        {
          "item": "Source freshness rule is documented",
          "status": "Yes / No / Partial",
          "guidance": "Define how stale sources are flagged, excluded, or escalated."
        },
        {
          "item": "Retrieval lineage is logged where needed",
          "status": "Yes / No / Partial",
          "guidance": "For material decisions, log which sources influenced the output."
        }
      ]
    },
    {
      "section": "6. Tool, integration, and system access",
      "items": [
        {
          "item": "Approved tools and systems are listed",
          "status": "Yes / No / Partial",
          "guidance": "Every integration is a permission boundary."
        },
        {
          "item": "Tool permissions are least-privilege",
          "status": "Yes / No / Partial",
          "guidance": "The agent should have only the access needed for approved tasks."
        },
        {
          "item": "System updates require explicit approval rules",
          "status": "Yes / No / Partial",
          "guidance": "Updating records is different from reading records."
        },
        {
          "item": "Integration failures have a fallback path",
          "status": "Yes / No / Partial",
          "guidance": "Define what happens when APIs, tools, data, or workflows fail."
        }
      ]
    },
    {
      "section": "7. Controls, evidence, and auditability",
      "items": [
        {
          "item": "Preventive, detective, and corrective controls are mapped",
          "status": "Yes / No / Partial",
          "guidance": "Define guardrails, logs, monitoring, approvals, alerts, and remediation."
        },
        {
          "item": "Prompt, context, output, and tool-call logging is defined",
          "status": "Yes / No / Partial",
          "guidance": "Log enough to investigate failures without over-collecting sensitive data."
        },
        {
          "item": "Evidence requirements match impact tier",
          "status": "Yes / No / Partial",
          "guidance": "Critical agents need stronger evidence and replayability."
        },
        {
          "item": "Kill switch, pause rule, or rollback owner is documented",
          "status": "Yes / No / Partial",
          "guidance": "Someone must be able to stop the agent fast."
        }
      ]
    },
    {
      "section": "8. Monitoring, incidents, and drift",
      "items": [
        {
          "item": "Monitoring signals are defined",
          "status": "Yes / No / Partial",
          "guidance": "Track accuracy, drift, incidents, overrides, misuse, latency, value, feedback, and control failures."
        },
        {
          "item": "Incident thresholds are documented",
          "status": "Yes / No / Partial",
          "guidance": "Define when agent behavior becomes reportable, escalated, restricted, or paused."
        },
        {
          "item": "Access and permission review cadence is defined",
          "status": "Yes / No / Partial",
          "guidance": "Tool access should not become permanent by default."
        },
        {
          "item": "Agent performance is reviewed against business value",
          "status": "Yes / No / Partial",
          "guidance": "A working agent that does not create value should be retired or redesigned."
        }
      ]
    },
    {
      "section": "9. Lifecycle and change governance",
      "items": [
        {
          "item": "Launch gate is documented",
          "status": "Yes / No / Partial",
          "guidance": "Define what is required for pilot, production, scale, and enterprise release."
        },
        {
          "item": "Change triggers are documented",
          "status": "Yes / No / Partial",
          "guidance": "Review on source, model, vendor, tool, workflow, policy, risk, or ownership changes."
        },
        {
          "item": "Retirement and supersession rules are defined",
          "status": "Yes / No / Partial",
          "guidance": "Agents must be removed when stale, unsafe, redundant, low-value, or replaced."
        },
        {
          "item": "Agent accountability is reviewed on cadence",
          "status": "Yes / No / Partial",
          "guidance": "Accountability must stay current as the organization, tools, and workflows change."
        }
      ]
    }
  ],
  "scale_versions": [
    {
      "company_size": "500+ employee company",
      "dimensions": [
        {
          "dimension": "Design intent",
          "recommended_pattern": "Create a lightweight agent accountability gate before teams connect AI to tools, knowledge, or workflows."
        },
        {
          "dimension": "Minimum scope",
          "recommended_pattern": "Track agent name, owner, purpose, approved tasks, blocked actions, approved sources, tool access, human review, logs, and lifecycle status."
        },
        {
          "dimension": "Operating pattern",
          "recommended_pattern": "Centralized AI owner or small review group approves agents that move beyond assistant-only use."
        },
        {
          "dimension": "AI focus",
          "recommended_pattern": "Prevent shadow agents, unmanaged prompt workflows, unclear owners, and agents with too much tool access."
        },
        {
          "dimension": "Governance need",
          "recommended_pattern": "Connect to AI governance checklist, ownership map, decision rights model, source-of-truth map, platform map, and control map."
        },
        {
          "dimension": "Red flags",
          "recommended_pattern": "A vendor feature is turned on without ownership, agents summarize stale knowledge, or teams treat agent action as a productivity shortcut without controls."
        }
      ]
    },
    {
      "company_size": "5,000+ employee company",
      "dimensions": [
        {
          "dimension": "Design intent",
          "recommended_pattern": "Create federated agent accountability across functions, platforms, data domains, and shared services."
        },
        {
          "dimension": "Minimum scope",
          "recommended_pattern": "Add impact tiering, tool permissions, integration ownership, logs, source lineage, incident thresholds, escalation, and control coverage."
        },
        {
          "dimension": "Operating pattern",
          "recommended_pattern": "Business unit owners register agents through a common template while enterprise governance routes higher-risk agents to security, privacy, data, legal, risk, and architecture."
        },
        {
          "dimension": "AI focus",
          "recommended_pattern": "Prevent duplicate agents, inconsistent review rules, unowned retrieval sources, and agents acting across systems without traceability."
        },
        {
          "dimension": "Governance need",
          "recommended_pattern": "Connect to integration map, data lineage map, evidence checklist, risk acceptance register, decision log, and workflow inventory."
        },
        {
          "dimension": "Red flags",
          "recommended_pattern": "Agents are embedded into operational workflows without system-of-record rules, clear human approval, or a stop path."
        }
      ]
    },
    {
      "company_size": "10,000+ employee company",
      "dimensions": [
        {
          "dimension": "Design intent",
          "recommended_pattern": "Create enterprise-grade agent accountability across business units, regions, vendors, regulated workflows, shared platforms, and multi-agent chains."
        },
        {
          "dimension": "Minimum scope",
          "recommended_pattern": "Full lineage across agent, owner, task, source, tool, integration, decision, control, risk, evidence, incident, value, and lifecycle state."
        },
        {
          "dimension": "Operating pattern",
          "recommended_pattern": "Central AI governance defines standards and risk tiers while federated owners manage registration, evidence, monitoring, controls, access review, and lifecycle governance."
        },
        {
          "dimension": "AI focus",
          "recommended_pattern": "Prioritize system-action agents, external communication agents, employee or customer-impacting agents, regulated workflow agents, and agents with write access."
        },
        {
          "dimension": "Governance need",
          "recommended_pattern": "Integrate with GRC, IAM, data governance, model risk, vendor risk, architecture, security, privacy, internal audit, and executive control-plane reporting."
        },
        {
          "dimension": "Red flags",
          "recommended_pattern": "Multi-agent workflows make it unclear who acted, what source was used, what tool was called, and who approved the outcome."
        }
      ]
    }
  ],
  "scoring_logic": [
    {
      "dimension": "Human ownership",
      "score_range": "0-5",
      "good_state": "One accountable business owner and supporting technical, data, decision, and control owners are clear."
    },
    {
      "dimension": "Scope clarity",
      "score_range": "0-5",
      "good_state": "Approved tasks, blocked tasks, start conditions, stop conditions, and handoffs are documented."
    },
    {
      "dimension": "Action boundary",
      "score_range": "0-5",
      "good_state": "Allowed and blocked agent actions are explicit and tied to decision rights."
    },
    {
      "dimension": "Data and knowledge boundary",
      "score_range": "0-5",
      "good_state": "Approved sources, blocked sources, source freshness, sensitivity, and retrieval lineage are clear."
    },
    {
      "dimension": "Tool access control",
      "score_range": "0-5",
      "good_state": "Tools, APIs, integrations, permissions, and update authority are least-privilege and owned."
    },
    {
      "dimension": "Human review design",
      "score_range": "0-5",
      "good_state": "Review triggers, reviewer role, evidence, approval authority, override path, and exceptions are clear."
    },
    {
      "dimension": "Control coverage",
      "score_range": "0-5",
      "good_state": "Preventive, detective, corrective, access, security, privacy, operational, and AI controls are mapped."
    },
    {
      "dimension": "Audit evidence",
      "score_range": "0-5",
      "good_state": "Prompts, context, sources, outputs, tool calls, approvals, exceptions, and actions are logged as needed."
    },
    {
      "dimension": "Monitoring readiness",
      "score_range": "0-5",
      "good_state": "Accuracy, drift, incidents, usage, overrides, value, feedback, and control failures are monitored."
    },
    {
      "dimension": "Lifecycle discipline",
      "score_range": "0-5",
      "good_state": "Launch, change, access review, pause, rollback, retirement, and supersession rules are documented."
    }
  ],
  "readiness_classification": {
    "0-1.9": "Unaccountable agent activity",
    "2.0-3.4": "Documented but weak agent accountability",
    "3.5-4.4": "Governed agent operating model",
    "4.5-5.0": "Enterprise-ready agent accountability"
  },
  "ai_prompts": [
    "Classify this agent by type, impact tier, strongest allowed action, affected systems, required owners, evidence needs, and governance route.",
    "Review this agent accountability checklist and identify missing owners, weak action boundaries, unclear source rules, excessive tool access, missing controls, and insufficient human review.",
    "Given this agent description, generate approved tasks, blocked tasks, approved tools, blocked tools, allowed actions, blocked actions, human review rules, and escalation path.",
    "Score this agent from 0 to 5 across ownership, scope clarity, action boundary, data boundary, tool access, human review, control coverage, audit evidence, monitoring readiness, and lifecycle discipline.",
    "Convert this agent accountability checklist into Lapemo objects for owner, agent, task, decision, source, platform, integration, control, evidence, risk, incident, and monitoring signal.",
    "Determine whether this agent should proceed locally, require domain review, require enterprise governance, require executive approval, require risk acceptance, or be blocked."
  ],
  "validation_rules": [
    "Every agent must have one accountable business owner before pilot, launch, or connection to enterprise tools.",
    "Every agent must be classified by type and strongest allowed action.",
    "Every agent must define approved tasks, blocked tasks, allowed actions, and blocked actions.",
    "Every agent must define approved sources, blocked sources, approved tools, and blocked tools.",
    "Agents with write access, external communication, customer impact, employee impact, regulated workflow impact, financial impact, security impact, privacy impact, or control impact must have documented human review and control coverage.",
    "Every moderate, high, critical, or regulated agent must have a kill switch, pause owner, rollback path, and escalation path.",
    "Any residual risk must link to risk acceptance or be explicitly marked as no acceptance required.",
    "Agents may not approve, commit, externalize, update systems of record, bypass controls, or make regulated decisions unless those actions are explicitly approved by governance.",
    "Agent logs must be sufficient to investigate prompts, context, retrieved sources, outputs, tool calls, approvals, exceptions, and actions for the assigned impact tier.",
    "Production agents must have review cadence, access review, change triggers, lifecycle status, and retirement or supersession rules.",
    "Status must never be encoded only by color. Use labels such as idea, design, pilot, production, scaled, restricted, paused, retired, or superseded."
  ],
  "lapemo_mapping": [
    {
      "object": "Agent object",
      "fields_entities": "Agent ID, name, type, purpose, lifecycle status, impact tier",
      "use": "Creates the canonical agent record."
    },
    {
      "object": "Ownership object",
      "fields_entities": "Business owner, technical owner, data owner, control owner, decision owner",
      "use": "Connects agent accountability to human ownership."
    },
    {
      "object": "Decision object",
      "fields_entities": "Decision influenced, allowed action, blocked action, approval owner, review rule",
      "use": "Prevents agents from becoming hidden decision makers."
    },
    {
      "object": "Information object",
      "fields_entities": "Approved sources, blocked sources, freshness, lineage, source owner",
      "use": "Controls what the agent can know and retrieve."
    },
    {
      "object": "Platform object",
      "fields_entities": "Approved tools, APIs, systems, permissions, system owner",
      "use": "Controls what the agent can access or change."
    },
    {
      "object": "Integration object",
      "fields_entities": "Tool calls, triggers, updates, data flows, failure paths",
      "use": "Maps agent-to-system dependencies."
    },
    {
      "object": "Control object",
      "fields_entities": "Preventive, detective, corrective controls, logs, thresholds, tests",
      "use": "Ties agent activity to governance controls."
    },
    {
      "object": "Evidence object",
      "fields_entities": "Prompt logs, source logs, output logs, tool calls, approvals, incidents, validation tests",
      "use": "Creates auditability and replayability."
    },
    {
      "object": "Risk object",
      "fields_entities": "Impact tier, residual risk, risk acceptance, mitigation, escalation",
      "use": "Routes risk and exception handling."
    },
    {
      "object": "Monitoring object",
      "fields_entities": "Accuracy, drift, incidents, overrides, adoption, value, control failures, access exceptions",
      "use": "Supports ongoing operational health."
    }
  ],
  "render_targets": [
    "DOCX",
    "PDF",
    "Markdown",
    "JSON",
    "Website page",
    "Interactive form",
    "Lapemo guided skill"
  ],
  "update_policy": {
    "review_cadence": "Quarterly by default, faster for agents with high-impact actions, write access, sensitive sources, production workflow use, incidents, vendor changes, model changes, or source changes.",
    "human_approval_required": true,
    "auto_update_behavior": "AI may recommend updates, flag stale ownership, detect access drift, identify missing controls, and propose revised boundaries. Human owner approves the new version."
  }
}