In this briefing
- 01ServiceNow connects a chat entry point to existing business rules
- 02Google Cloud and ServiceNow separate industry actions by delivery status
- 03Salesforce connects evaluation results to intervention actions
- 04Cross-platform control still depends on identity, asset catalogues and execution boundaries
- →What to watch next
- ↗Sources and verification
- The ServiceNow plugin is available in the OpenAI marketplace. ServiceNow Action Fabric passes requests from ChatGPT to existing business rules, permissions and approval processes, while ServiceNow remains responsible for executing and governing the associated actions.
- ServiceNow and Google Cloud have marked the retail, customer-relationship-management and information-technology workflows as generally available, but the telecommunications workflow is still planned for general availability in early 2027. A shared chat entry point cannot remove differences in delivery status between business processes.
- Salesforce already provides the base Agent Kill Switch, and Salesforce plans to introduce Agent Evals, API Evals and evaluation-triggered intervention capabilities in stages between November 2026 (month 11) and January 2027 (month 1). A gateway block still may not terminate every process on an external platform.
ServiceNow connects a chat entry point to existing business rules
On 28 October, OpenAI and ServiceNow announced an expanded partnership. The new ServiceNow plugin allows a user to find similar incidents, assess change risk and update ServiceNow records from ChatGPT. ServiceNow Action Fabric applies the organisation's existing business rules, permissions and approvals, while ServiceNow remains the system that executes and governs those actions. OpenAI also states that a customer can choose which ServiceNow capabilities the customer makes available to ChatGPT.
The two companies list current capabilities separately from future capabilities. The ServiceNow plugin is available in the OpenAI marketplace, and enhanced capabilities are expected by the end of 2026. Voice capability for ServiceNow Super Intelligence Control Tower is expected in the first quarter of 2027. OpenAI-supported security capability is expected to enter early access in January 2027 (month 1) and become generally available in the second quarter of 2027. Integration between Codex and ServiceNow AI Workflow Factory is expected during 2027.
For enterprise adoption, a chat entry point can produce a controlled action only when the entry point calls existing business objects, permissions and approval processes. An enterprise needs to convert a model-generated request into an explicit actor, target record, normalised parameters and approval state, and the enterprise needs to retain the execution result that ServiceNow returns. The currently available plugin can provide a connection point, but an enterprise cannot place future security, voice and development capabilities within the current acceptance scope.
Google Cloud and ServiceNow separate industry actions by delivery status
On 27 October, ServiceNow announced industry-workflow integrations with Google Cloud. The retail-store-operations integration can report and diagnose equipment issues and dispatch a field technician. The customer-relationship-management integration can update opportunity records, create follow-up work and produce configuration and quotation information. The information-technology service-and-operations-management integration can process incidents, change risk and configuration dependencies. ServiceNow marks these three types of integration as generally available to joint customers.
The telecommunications network-inventory and service-order-management integration is still expected to become generally available in early 2027. ServiceNow Action Fabric uses Model Context Protocol servers and Agent2Agent connections to make workflows available in Gemini Enterprise, while ServiceNow plans to publish connectors, demonstrations and reference architectures through Google Cloud Marketplace. The announcement does not provide cross-industry error rates, complete service levels, independent performance results or a list of all regions.
For enterprise adoption, a connection protocol can unify an invocation entry point, but the business meaning still comes from specific records, processes, permissions and data relationships. An enterprise needs to verify separately the identifiers, state changes and rollback rules for objects such as retail equipment, customer opportunities, information-technology configuration items and telecommunications orders. An enterprise knowledge graph or ontology can help a system align object relationships and business definitions, but the enterprise still needs to prove each action through the target system's permissions and execution result.
Salesforce connects evaluation results to intervention actions
On 27 October, Salesforce announced a new timetable for Agent Fabric. Agent Evals is planned to generate tests before deployment and to assess Agents, models and Model Context Protocol servers against quality thresholds during production operation. API Evals is planned to generate function and performance tests from application programming interface specifications. Both capabilities are planned to enter testing in November 2026 (month 11) and become generally available in January 2027 (month 1).
Salesforce lists the base Agent Kill Switch as generally available, while Salesforce lists the evaluation-triggered Agent Kill Switch as generally available in January 2027 (month 1). VentureBeat further notes that the retail example in the Salesforce announcement is a hypothetical scenario and that blocking governed traffic through a gateway may not terminate every process that continues to run on an external platform. The Salesforce Custom Agent Scanner is in testing, while Open Policies for Federated API Gateways is generally available.
For enterprise adoption, after a system detects a wrong answer, an interface change or a cost anomaly, the system still needs to map the detection result to an alert, approval, rate limit, isolation or recovery action. An enterprise needs to define whether an intervention affects one request, a session, an Agent, a tenant or an external running process, and the enterprise needs to verify recovery authorisation and the audit record. A product roadmap can show that a control layer is reducing the distance between detecting a problem and executing an intervention, but the product roadmap cannot prove that automatic intervention already operates safely in the target business process.
Cross-platform control still depends on identity, asset catalogues and execution boundaries
On 28 October, ServiceNow also announced the Control Tower Alliance. The programme's founding members include Anthropic, Amazon Web Services, Cisco, HiddenLayer, Kong, NVIDIA, Okta, OpenAI, Stripe, WitnessAI and Zenity. ServiceNow states that integration solutions entering the programme will undergo qualification, certification and integration testing. The current announcement does not list each specific solution that has passed, the test methods, failed results or a shared service level.
The ServiceNow announcement also describes cross-platform stopping as a future direction. Okta's statement explains that the parties' current work aims to prevent anomalous Agents from obtaining new tokens, and that the parties want to advance a cross-platform stopping standard. The Salesforce announcement places internal environments, third-party platforms and different application programming interface gateways within the scope of asset discovery and policy. The two companies describe their respective products and partnership roadmaps, while the current material does not prove that the two control layers already interoperate.
For enterprise adoption, a control layer can implement a policy accurately only when the system can identify the Agent identity, model, tool, business object, application programming interface and execution location. An enterprise needs to establish a unified asset catalogue and related semantics, and the enterprise needs to retain separate evidence for identity revocation, gateway blocking, external-process termination and business rollback. A partnership alliance and a cross-platform vision can reduce future integration fragmentation, but an enterprise cannot treat a membership list or a planned standard as completed interoperability verification.
What to watch next
- Can OpenAI and ServiceNow publish customer-verifiable permission mappings, approval records, failure modes, service levels and regional coverage for the current plugin?
- Can ServiceNow and Google Cloud use real customer workloads to publish success rates, rollback outcomes, human-approval proportions and complete task costs for different industry workflows?
- Can ServiceNow, Salesforce, Okta and other partners prove that identity revocation, gateway blocking, external-process termination and business rollback remain consistent across platforms?
Sources and verification
Golden Data has edited this briefing from the public materials listed above. The original sources govern facts and figures. The enterprise relevance sections are Golden Data editorial analysis and do not constitute an endorsement of any third-party product.
← Back to AI Daily Briefing