ServiceNow AI Agents

ServiceNowのAI Agent Studioを、AI agent/agentic workflowの作成・管理・テスト・監視・AI Guardian・analytics・release/role依存の組織overlayとして更新する。

organization overlay / provider mapping / governance / benchmark / security / audit / operability。Now Platform・Otto AI Agents・Virtual Agentを別availability gateで扱う。

ServiceNow公式Docsの現行releaseページがAI Agent Studioの作成・実行ログ・手動/自動評価・security control・analytics・前提インストールを何と記載しているか。

• source_class=official product documentation; exact_title=AI Agent Studio overview; author/org=ServiceNow; canonical_url=https://www.servicenow.com/docs/r/intelligent-experiences/ai-agent-studio.html publication/update/access_date=Release version Australia / Updated 2026-03-12 / accessed 2026-09-13 JST; version/release/tag/commit/API revision=Australia release; supporting_locator=Overview lines 2043-2072（centralized applicationでAI agents/agentic workflowsをcreate/manage/test/monitor、Otto AI Agentsのinstall、execution logs、version filter、manual single execution/automated multi-execution evaluations、AI Guardianのoffensiveness/prompt injection/long-term memory、analytics）; evidence_label=VERIFIED_SPEC+PRODUCT_VOLATILE; relation=support+limit
• source_class=official product documentation; exact_title=AI Agent Studio overview; author/org=ServiceNow; canonical_url=https://www.servicenow.com/docs/r/intelligent-experiences/ai-agent-studio.html publication/update/access_date=Australia release / accessed 2026-09-13 JST; version/release/tag/commit/API revision=Australia release; supporting_locator=Testing agentic AI lines 2099-2105（manual executionとautomated evaluations、security-control access test、live interactionはVirtual Agentを使う必要があるとの制限）; evidence_label=VERIFIED_SPEC+PRODUCT_VOLATILE; relation=support+limit

• ESTABLISHED: ServiceNow公式DocsはAI Agent Studioをagent/agentic workflowの作成・管理・テスト・監視アプリとして記載し、version付きexecution logs、manual/automated test・evaluation、AI Guardian設定、analyticsを案内している。全機能にはServiceNow Otto AI Agentsのinstallが必要と記載し、live agent interactionのtestingはVirtual Agentを使う制限も明示している。
• NOT_ESTABLISHED: Australia releaseのDocs記述は、他release・全tenant・全plan・全regionでの機能有効化、AI Guardianが全prompt injectionを阻止すること、analyticsが因果的ROIや安全率を証明すること、実業務での自律action安全性を確立しない。

release差・plugin/install差・role/ACL・Virtual Agent session制限、評価データの代表性、AI Guardian false positive/negative、agent loop・trigger recursion・tool side effect、analyticsの計測欠損がある。Docsページ自身のAI-generated summaryはas-isで完全性保証なしと注記されているため、本文記載を優先する。

ServiceNow release、AI Agent Studio/AI agents/AI Guardian/Virtual Agentのinstall・license・subscription、tenant/region、admin role・ACL・scope、Now Assist/LLM provider、workflow/table/data permissions、execution/evaluation quotas、data retention。

ServiceNow official docs + PRODUCT_VOLATILE。Australia releaseの公開仕様確認のみ。tenant activation、role付与、agent作成、live interaction、外部書込み、課金・本番評価は未実施。

Docsはmanual test、automated multi-execution evaluations、execution logs、analytics metricsを仕様として記載するが、specific dataset・ground truth・threshold・human agreement・業務品質改善値は本Packetで確立していない。UNKNOWN。

Security=AI Guardian、ACL、role、approval、tool scope、Virtual Agent境界をactionごとに確認。Privacy=Now tables、prompt、memory、logs、LLM providerのdata policy。Audit=release、agent/workflow version、execution、test/eval、policy/ACL、approverを相関。Cost=agent execution、model、tool/Assist consumption、evaluation、storageを含む。Operability=release upgrade、session timeout、loop kill switch、rollback、evaluation regression、log completenessを試験。

「ServiceNow AI Agent Studioの公式Docs（Australia release、2026-03-12更新）は、AI agent/agentic workflowの作成・管理・テスト・監視、execution log、manual/automated evaluation、AI Guardian、analyticsを記載している。Otto AI Agentsのinstall等の前提があり、release・tenant・role・license・session境界と実業務安全性は別途検証が必要である。」

ServiceNow organization overlayにrelease・install・role・ACL・Virtual Agent制約を引き継ぐ。次回は各releaseのfeature matrix、license/pricing、AI Guardian仕様、evaluation schema、audit/data retention、公開incident/failure report、loop/trigger/Assist costの再現可能TEST_PROFILEを補う。

current release beyond Australia、exact licensing/region、AI Guardian detection metrics、evaluation ground truth、agent loop/recursion controls、ACL/data boundary、Assist cost、tenant availability、公開incident。

Australia Docsの存在をユーザーtenantの有効化・全release・全機能保証へ外挿しない。analytics・manual test・AI Guardianの存在を安全率・ROI・prompt injection完全防御と書かない。 UNIT_CLOSE_STATUS: INDIVIDUAL_BOUNDED_CLOSE / EVIDENCE_RECORDED / UNKNOWN_PRESERVED

===== RUN CLOSE / HANDOFF ===== INDIVIDUAL_PACKET_COUNT: 9/9（EA-137..EA-145）

11 source records across 9 packets; official project/product documentation, author-maintained repositories, and one primary research paper record exact title, author/org, canonical URL, access/publication date, version/release/tag/API status, supporting locator, evidence label, and relation.

EA-137..145 each has separate question, established/unestablished findings, evidence relation, limits/counterevidence/failure envelope, availability gates, implementation state, benchmark/eval, safe wording, handoff, open unknown, and stop condition.

Framework/API/docs establish documented concepts and current page claims only. Historical paper results, provider marketing, customer quotes, product existence, plan/region/org/role availability, tenant behavior, production safety, and business outcomes remain separately gated or UNKNOWN. LIVE_SIDE_EFFECTS: 0. No authenticated tenant operation, external write, billing, contract change, secret input, live agent execution, or dangerous test.

ServiceNow official docs still document AI Agent Studio/manual and automated evaluations; 2026-08-14 Zurich AI Guardian docs require admin enablement and describe mitigation, not complete prevention. Australia evaluation docs require agent versions, datasets, permissions and execution logs. Zurich Patch 4 may turn designated AI assets on by default only under stated install/admin conditions.

ServiceNow documents AI Agent Studio for creating/managing/testing agentic assets, execution/evaluation workflows, guardrails and monitoring. Availability and behavior remain release/plugin/license/role/ACL/tenant dependent.

independent production success/safety rates; Guardian false-positive/false-negative rates; exact tenant entitlement; complete cost/Assist consumption; organization-specific data boundary.