ServiceNow Third-Party Model Defaults for Otto / Now Assist: A UK Control Tower Playbook

Third-party defaults are a store-app gate — not a platform surprise

From 9 July 2026, ServiceNow began rolling out third-party model providers as the default for out-of-the-box (OOTB) Now Assist skills and AI agents. The change is not a single-night cutover. It is tied to each product's store app release schedule: when you apply that product's update, OOTB skills and agents whose default provider you never changed switch to a ServiceNow-validated third-party provider.

For UK platform owners, AI Stewards, CAB chairs and DPOs, this is a control-tower and change-window moment. Azure OpenAI, AWS Claude and Google Gemini were already supported on the ServiceNow AI Platform. What is new is that OOTB defaults move automatically with store updates unless you already set an explicit instance-, skill-, skill-group- or agent-level provider.

Otto is the renaming path for Now Assist experiences. In this playbook we say Otto / Now Assist where the product UI still shows either label — treat them as the same governance surface for model providers, Assists and Control Tower policy.

Enterprise automation and AI control

What ServiceNow is changing — and what it is not

ServiceNow's rationale is continuous improvement: frontier models improve reasoning and task completion quickly, and product teams validate OOTB skills against the assigned provider before it becomes the default. That is useful for UK estates — if, and only if, your allowed-provider policy and DPIA language already cover the destination.

Two timelines you must keep separate

Timeline A — OOTB default switch (from 9 Jul 2026)

When you take a product's store app update, skills/agents still on OOTB Now LLM defaults move to the validated third-party default for that skill. Check assignments after update in:

• AI Control Tower → Configurations → Controls → AI Model Providers (Impact Summary)
• Now Assist Admin Console → Settings → Manage AI Models → Manage Model Providers

Timeline B — Now LLM Service commercial retirement (later)

Now LLM Service general-purpose models for commercial deployments are on a separate deprecation/retirement path. Community guidance commonly cites a December 2026 retirement target for commercial Now LLM use — always confirm the live model status table in the Now Assist Model Lifecycle KB before you brief the board.

Critical nuance for CAB:

• Explicit Now LLM configs are not overwritten by the July 2026 store updates.
• The forcing function arrives when Now LLM models are retired.
• Custom skills built on Now LLM need clone → retest → evaluate on a third-party provider before retirement, not after the notification.

UK pre-update gate (do this before every store app)

1. Allowed providers in AI Control Tower

By default all providers may be available. Many UK regulated estates have restricted the list.

Failure mode: you restrict AWS Claude (or Gemini / Azure OpenAI); a store update assigns that provider as the new OOTB default; fallback is disabled → the skill/agent is blocked by your own policy.

Gate:

1. Open AI Control Tower → Configurations → Controls → AI Model Providers.
2. Export or screenshot the allowed set and owners.
3. Review Impact Summary for skills/agents that would be affected before promoting the store app.
4. Decide: expand allowed providers or set an explicit allowed provider on the skill/agent.

2. Configuration scenario matrix

3. Sub-production order of operations

1. Apply the store app update in sub-production only.
2. Open Impact Summary + Manage Model Providers; record old vs new provider per skill/agent.
3. Run dual evaluation on top business-critical OOTB skills (ITSM summarisation, catalog, HR/CSM if licensed).
4. Update CAB / AI change pack with provider deltas and residual risk.
5. Promote to production in a controlled window — per product, matching the staggered schedule.

GDPR, DPIA and UK ops language

Third-party defaults do not invent a new data path you never discussed — but boards and DPOs will ask. Keep the evidence pack boring and complete:

If your DPO previously approved "Now LLM only", treat the default switch as a material control change even when Assists and OOTB workflows stay the same.

Custom skills — the quiet retirement risk

July store updates do not rewrite custom skill providers. Waiting until Now LLM retirement to retune prompts is how UK programmes create December fire drills.

Playbook for developers / AI Stewards:

1. Inventory custom skills in Now Assist Skill Kit still on Now LLM.
2. For each: clone the prompt; point the clone at an approved third-party provider.
3. Run human + automated evaluations; capture regression notes (tone, grounding, PII redaction behaviour).
4. Publish the validated replacement before the retirement target in the model status table.
5. Keep the old skill in a named rollback set until dual-run confidence is signed off.

Prompt tuning across providers takes calendar time. Work backwards from the retirement target, not forwards from the first warning email.

Air-gapped, self-hosted and regulated-sovereign estates

This transition describes commercial deployments with connectivity to ServiceNow-integrated third-party providers. It does not apply yet if you are:

• Air-gapped / isolated (no third-party provider connectivity)
• Self-hosted / on-prem where third-party access is not in the deployment model
• Regulated or sovereign markets where residency limits third-party availability

Those estates continue on Now LLM Service for now. ServiceNow has stated an open-weights hosted model package is in development for these deployment types, with availability and migration guidance to follow. Isolated Markets / Self-Hosted customers should stay close to their account team — do not force a commercial third-party default playbook onto an estate that cannot reach those endpoints.

CAB checklist (print this)

1. Allowed providers reviewed in AI Control Tower before any Now Assist / Otto store app.
2. Impact Summary owned by named AI Steward + platform owner.
3. Explicit configs inventoried (instance / skill / agent) — especially Now LLM holdouts.
4. Sub-prod dual-run evidence attached for top 5 business skills.
5. DPIA / AI inventory updated for provider defaults (Assists unchanged noted).
6. Custom Now LLM skills on a dated migration plan toward Dec 2026 commercial retirement window (confirm KB dates).
7. Air-gapped / sovereign path confirmed out of scope or escalated to account team.
8. Production promotion scheduled per product, not as a mega-release.

Risks if you skip the gate

Closing

Third-party defaults for OOTB Otto / Now Assist skills are a store-app and Control Tower programme, not a surprise model rewrite. Explicit configs stay; Assists and platform data protections stay; air-gapped and sovereign paths stay on a different track. What changes for UK production is the discipline of allowed providers, Impact Summary review, dual-run evidence and a Now LLM retirement plan that starts before the calendar forces it.

If you want a structured AI Control Tower / model-provider readiness pass against your UK ServiceNow estate — including CAB language and custom-skill migration sequencing — AIATS offers a Free Evaluation: practical, UK-enterprise, no theatre.

Questions for the AI Steward / CAB agenda

1. Which providers are allowed today, and who can change that list?
2. Which store apps in the next 30 days carry Now Assist / Otto model-default changes?
3. How many skills/agents still sit on explicit Now LLM?
4. What is our dual-run standard before promoting a provider switch?
5. Is the DPIA / AI inventory updated for Azure OpenAI, AWS Claude and Google Gemini via ServiceNow?
6. Are air-gapped or sovereign instances correctly excluded from this playbook?
7. Who owns the custom-skill clone/evaluate backlog before Now LLM commercial retirement?