sources · source-BATCH-2026-005-009
Practices for Governing Agentic AI Systems
Accepted discovery candidate candidate-BATCH-2026-001-085; contributes company white paper and policy analysis evidence or normative context to governed AI-agent identity.
Open the canonical original source
Source at a glance
- Source type
- research report
- Publisher
- OpenAI
- Source or access date
- 2026-08-15
- Analysis depth
- deeply analyzed
- Rights treatment
- link and paraphrase
- Human review
- Murray Newlands · 2026-08-16
What the source reports or argues
- The authors propose evaluation, action-space constraints, default behaviors, legibility, monitoring, attributability, and interruptibility as building blocks.
- They argue that at least one human legal entity should remain accountable for uncompensated direct harm.
Important limitations
- No empirical outcome evaluation.
- Company-published and funding/conflicts are not reported.
- Many recommendations are explicitly preliminary.
Source-located statements
8 reviewed statements are indexed from this source.
- The paper treats agenticness as a matter of degree across goal complexity, environmental complexity, adaptability, and independent execution. (Official PDF p. 4, Section 2.1 "Agenticness, Agentic AI Systems, and Agents", 2023 version)
- The paper separates three principal lifecycle roles: model developer, system deployer, and user. (Official PDF pp. 5-6, Section 2.2 "The Human Parties in the AI Agent Life-cycle", 2023 version)
- The authors argue that every uncompensated direct harm caused by an agentic system should remain attributable to at least one accountable human legal entity. (Official PDF p. 3, Section 1 "Introduction", 2023 version)
- Deployers or users should test whether an agent is suitable for its intended task under conditions resembling deployment. (Official PDF pp. 8-9, Section 4.1 "Evaluating Suitability for the Task", 2023 version)
- High-consequence actions should be withheld from agents, bounded by hard limits, or made contingent on informed human approval. (Official PDF pp. 9-10, Section 4.2 "Constraining the Action-Space and Requiring Approval", 2023 version)
- Users or deployers can place a separate automated monitor over an agent's activity when human review cannot match its speed or volume. (Official PDF pp. 12-13, Section 4.5 "Automatic Monitoring", 2023 version)
- For high-stakes exchanges, a counterparty could require an agent identifier connected to its human principal and accountability information. (Official PDF pp. 13-14, Section 4.6 "Attributability", 2023 version)
- A user's shutdown authority should cover both the primary agent and any subagents it initiated. (Official PDF pp. 14-15, Section 4.7 "Interruptibility and Maintaining Control", 2023 version)
Evidence lineage and transparency
| Relationship field | Linked identifiers |
|---|---|
| author ids | person-BATCH-2026-005-yonadav-shavit, person-BATCH-2026-005-sandhini-agarwal, person-BATCH-2026-005-miles-brundage, person-BATCH-2026-005-steven-adler, person-BATCH-2026-005-cullen-o-keefe, person-BATCH-2026-005-rosie-campbell, person-BATCH-2026-005-teddy-lee, person-BATCH-2026-005-pamela-mishkin, person-BATCH-2026-005-tyna-eloundou, person-BATCH-2026-005-alan-hickey, person-BATCH-2026-005-katarina-slama, person-BATCH-2026-005-lama-ahmad, person-BATCH-2026-005-paul-mcmillan, person-BATCH-2026-005-alex-beutel, person-BATCH-2026-005-alexandre-passos, person-BATCH-2026-005-david-g-robinson |
Machine review: machine checked. Human review: approved. Workflow: published.
Complete structured record
- source id
- source-BATCH-2026-005-009
- canonical title
- Practices for Governing Agentic AI Systems
- alternate titles
- source type
- research_report
- series or parent source
- official OpenAI PDF
- publisher
- OpenAI
- channel
- Unknown
- speaker ids
- author ids
- person-BATCH-2026-005-yonadav-shavit, person-BATCH-2026-005-sandhini-agarwal, person-BATCH-2026-005-miles-brundage, person-BATCH-2026-005-steven-adler, person-BATCH-2026-005-cullen-o-keefe, person-BATCH-2026-005-rosie-campbell, person-BATCH-2026-005-teddy-lee, person-BATCH-2026-005-pamela-mishkin, person-BATCH-2026-005-tyna-eloundou, person-BATCH-2026-005-alan-hickey, person-BATCH-2026-005-katarina-slama, person-BATCH-2026-005-lama-ahmad, person-BATCH-2026-005-paul-mcmillan, person-BATCH-2026-005-alex-beutel, person-BATCH-2026-005-alexandre-passos, person-BATCH-2026-005-david-g-robinson
- institutional author
- Unknown
- organization references
- recorded at
- Unknown
- event date
- Unknown
- published at
- Unknown
- updated at
- Unknown
- duration seconds
- Unknown
- language
- lang-en
- translated title
- Unknown
- translation method
- Unknown
- geography of speaker
- geography of organization
- geography discussed
- global policy context
- study geography
- global policy context
- original url
- https://cdn.openai.com/papers/practices-for-governing-agentic-ai-systems.pdf
- canonical url
- https://cdn.openai.com/papers/practices-for-governing-agentic-ai-systems.pdf
- archived url
- Unknown
- embed url
- Unknown
- doi
- Unknown
- canonical identity status
- verified
- repost status
- original_or_authoritative_rendition
- original source id
- Unknown
- rights status
- link_and_paraphrase
- ownership status
- third_party
- relationship to off
- none_identified; Executive AI Research snapshot contains no matching production records
- transcript status
- Unknown
- transcript source
- Unknown
- transcript republication permission
- Unknown
- chapter markers
- analysis basis
- Complete official 23-page PDF reviewed; definition, accountability, monitoring, attribution, interruptibility, indirect impacts, and acknowledgements inspected visually.
- topics
- topic-ai-agents, topic-ai-governance, topic-cybersecurity, topic-non-human-identity
- executive roles
- role-ciso, role-cio, role-cto, role-general-counsel, role-board-director
- original abstract
- Unknown
- inclusion rationale
- Accepted discovery candidate candidate-BATCH-2026-001-085; contributes company white paper and policy analysis evidence or normative context to governed AI-agent identity.
- source quality dimensions
- {"attribution_strength":"high","methodological_transparency":"appropriate_for_policy_analysis","independence":"company_published","bibliographic_stability":"high_official_pdf","evidence_strength":"moderate_normative_policy_analysis_not_empirical","unresolved":"Publication day, funding, affiliations, conflicts, and external review status are not reported in the PDF."}
- methodology quality
- {"design":"non-empirical company white paper and policy analysis","transparency":"appropriate_for_policy_analysis","human_review_required":true}
- study design
- non-empirical company white paper and policy analysis
- sample
- {"size":"not applicable","sampling_method":"not applicable"}
- population
- model developers, system deployers, users, and third parties in agentic-system lifecycles
- date range
- {"fieldwork":"not applicable","version":"2023; PDF metadata created 2023-12-18"}
- funding
- not reported
- sponsor
- OpenAI publisher; separate sponsor involvement not reported
- peer review status
- not reported; not treated as peer reviewed
- findings
- The authors propose evaluation, action-space constraints, default behaviors, legibility, monitoring, attributability, and interruptibility as building blocks., They argue that at least one human legal entity should remain accountable for uncompensated direct harm.
- limitations
- No empirical outcome evaluation., Company-published and funding/conflicts are not reported., Many recommendations are explicitly preliminary.
- correction ids
- retraction status
- none_identified_on_official_pdf_url_as_of_2026-08-15
- content hash
- 22b3a8607ed781a848b82b0bfe8e638b16cd027aac90a19b2aecf236063d0e7c
- accessed at
- 2026-08-15
- verification status
- metadata_content_and_version_machine_verified_human_approved
- source depth
- deeply_analyzed
- publication status
- published
- workflow status
- published
- machine review status
- machine_checked
- human review status
- approved
- reviewed by
- Murray Newlands
- reviewed at
- 2026-08-16T23:17:22Z
Provenance and revision history
{
"provenance": [
{
"source_url": "https://cdn.openai.com/papers/practices-for-governing-agentic-ai-systems.pdf",
"accessed_at": "2026-08-15",
"retrieval_method": "official source retrieval and machine-assisted review",
"exact_locator": "2023; PDF metadata created 2023-12-18",
"content_hash": "22b3a8607ed781a848b82b0bfe8e638b16cd027aac90a19b2aecf236063d0e7c",
"batch_id": "BATCH-2026-005",
"prompt_id": "OEII-EVIDENCE-RESEARCH",
"prompt_version": "2.0",
"notes": "Human review pending."
}
],
"revision_history": [
{
"changed_at": "2026-08-16T23:17:22Z",
"changed_by": "Murray Newlands",
"summary": "Approved for the governed-identities pilot release under the exact scope, exclusions, rights treatment, and limitations recorded in issue #18.",
"batch_id": "BATCH-2026-005"
}
]
}