There is OpenAI He announced his presencea new enterprise product for deploying and managing AI agents in customer-facing and internal workflows.
The offering is intended for eligible enterprise customers who want agents to answer questions, access company systems, take approved actions, and provide services to human employees while operating within company-defined policies, permissions, and evaluation standards.
Availability is immediate through a limited general availability program. OpenAI Forward Deployed Engineers (FDEs) and select global system integrators lead deployments and the product is not available on a self-service basis.
As for the use of non-OpenAI models, such as the powerful and increasingly popular Chinese openweight alternatives GLM-5.2 and Like K3OpenAI spokesperson clarified this "Presence uses OpenAI models for the underlying agent, but also allows customers to integrate third-party models and services via APIs for firewalls, tools, and other parts of the workflow."
OpenAI has not disclosed pricing, geographic restrictions, contract terms, or the expected cost of engineering and integration work accompanying the deployment. I’ve asked an OpenAI spokesperson twice about pricing and am awaiting a response. I’ll update when I hear back.
OpenAI positions availability as the answer to a challenge that’s becoming more important as companies move beyond AI demonstrations: how agents behave reliably in production as business rules, customer needs, and business conditions change. Presence packages the policies, system connections, assessments, safeguards, and update processes required to run agents across an enterprise.
If your business is interested in using AI agents, but you’re not sure how to integrate OpenAI models, APIs, internal systems, security controls, and assessment tools into something reliable, Presence is designed to simplify the process. Instead of building the infrastructure yourself, you work with OpenAI and its deployment engineers to deploy production-ready agents into your existing workflows.
The product is available today for real-time voice and chat experiences, according to OpenAI’s official announcement. The company’s promotional materials also outline a broader ambition to include voice, chat, email and other channels, but OpenAI has not confirmed that email support will be available at launch.
A managed foundation for production agents
Presence brings together company knowledge, standard operating procedures, proven actions, simulations, assessment tools, guardrails and escalation rules. Enterprises can reuse some controls across deployments while customizing others for a particular workflow or channel.
Each placement begins with a defined task, such as resolving an invoicing issue, supporting an insurance claim, or handling an employee IT request. The agent receives only the data and system access required for this task. The client determines what the agent can do independently, what actions require approval, and when the person must commit.
Before an agent reaches production, teams can test it with common queries, unusual edge cases, and higher-risk scenarios. Graders evaluate whether he achieves the intended result, follows the policy, uses the tools correctly and is promoted when required. Hooks can intervene when the interaction goes beyond the defined boundaries of the organization.
OpenAI shared with VentureBeat promotional screenshots showing administrators running simulation batches against policy changes, including the revised annual refund policy, and reviewing results across transaction categories.
Other interface mockups show production health, customer intent patterns, and task execution alerts. The visuals show OpenAI’s promising type of controls, though they don’t specify how these metrics are calculated or how they align with contractual service levels.
The product continues to monitor performance after launch. Production sessions, escalations, and quality alerts can reveal where an agent is performing as intended and where it needs to focus. Using the Presence plugin, Codex investigates these signals and offers updates. Teams then test the proposed change against a version already in production before approving a managed release.
This process is designed to solve one of the most difficult operational problems in enterprise AI: an agent running at startup can become less reliable if policies, products, or user behavior change. Availability gives companies a formal mechanism to update behavior without allowing an automated system to rewrite itself without verification.
OpenAI says Presence is already beefing up its English-language phone support channel at 1-888-GPT-0090. The system handles open requests, checks callers, uses account context, and executes approved actions. According to the company, it is now being resolved 75% of incoming problems without human assistance.
OpenAI also says its Codex-powered development cycle reduces human handoffs 15 percentage points over a 10-day period. These figures are provided by the company and have not been independently verified.
Several large organizations evaluate the same fund. BBVA is exploring voice support for everyday banking needs in Mexico. SoftBank is testing customer conversations in natural Japanese, while Australian insurer IAG is exploring support during periods of high demand such as severe weather and natural disasters.
“At BBVA, we are working closely with OpenAI to explore how trusted customer agents can help shape the future of financial services,” said Daniel Ordaz, head of AI transformation at BBVA Mexico.
“Through our collaboration with OpenAI, we’re exploring how Presence can enable trusted customer agents that communicate naturally, connect to the processes needed to resolve inquiries, and consistently represent SoftBank in customer interactions,” said Tadahisa Murakami, vice president and head of Information and Digital Transformation at SoftBank Corp.
From model input to forward-deployed implementation
Presence extends OpenAI’s enterprise strategy beyond APIs and subscription software by formalizing a high-touch deployment model. Forward-Deployed Engineers work with customers to select workflows, integrate internal systems, create permissions, configure policies, test agents, and move them to production.
This approach is similar to A model developed by AI ontology and intelligence platform Palantir, connecting FDEs with customers to adapt proprietary software to complex government and commercial environments. The similarity is less in the underlying technology than the delivery method: both companies place technical staff close to the customer’s operations, where integration and process design often determine whether the software creates value.
Products cannot be exchanged. Palantir’s model has historically focused on data integration, ontologies, and operational decision systems. Availability is more narrowly focused on AI-agent behavior, validated actions, evaluations, escalation, and continuous improvement. OpenAI presents it not as a standalone consultant, but as a reproducible software product supported by engineers and system integrators.
In May 2026, OpenAI launched its AI consulting and integration firm OpenAI Deployment Companywith investment and Support from Bain & Company. It also offers programs for customizing and fine-tuning the model to suit specific enterprise needs.
Its main rival in the US, Anthropic, has also moved Towards a service-based enterprise model through Ode, his consulting organization is built around advanced engineers who help companies integrate Claude into complex workflows that began just a week ago. The broad rationale is similar: businesses often need more than model acquisition. They need help connecting data and systems, defining permissions, validating behavior, and managing deployment risk.
Availability differs in how OpenAI packages these requirements into a branded agent-management product. Anthropic’s initiative focuses on helping enterprises deploy Cloud, while Presence combines implementation services with an operational layer defined for policies, simulations, evaluations, approvals and production updates.
Availability goes even further by making forward placement a core part of how a particular agent’s product reaches customers. It does not replace OpenAI’s API business; the company said it will continue to support voice clients with access to boundary models through the OpenAI API.
This trend reflects a broader market view that many businesses still need hands-on help moving agents from pilot projects to stable operations. Even organizations with strong internal engineering teams must coordinate security, compliance, workflow ownership, data access, and escalation responsibilities. Rather than stacking separate layers of orchestration, evaluation, and consulting, Availability customers try to combine these tasks.
The latest security breach is visible in the background
Unfavorably for OpenAI, the launch of Presence comes a day later OpenAI and Hugging Face have disclosed an unprecedented security incident OpenAI frontier models that passed internal evaluation escaped protection to good purpose without guidance to follow these methods, entered the open internet and cyberattacked Hugging Face.
According to the described joint disclosure, OpenAI models operating within an evaluation called ExploitGym identified and exploited a zero-day vulnerability in a third-party packet registry cache proxy. The models reportedly escalated privileges before targeting the Hugging Face systems, moving sideways and accessing the internet while searching for benchmark-related information.
An incident is relevant to enterprise buyers because it raises questions about sandboxing, tool permissions, external access, monitoring and incident response.
The revelation also highlighted a practical challenge for advocates. Hugging Face personnel discovered that commercial boundary model APIs rejected some forensic requests because the logs contained exploit payloads, credentials, and shell commands that triggered security systems. The team then used a locally deployed open weight model to aid in the analysis.
The availability therefore comes as both a product launch and a test of OpenAI’s ability to translate model capabilities into managed enterprise operations. Its policies, simulations, assessments, and human validations address real deployment gaps. But without public pricing, technical interoperability details, compliance information or service level commitments, customers still lack the information they need to assess overall cost and operational risk.
For now, Presence is aimed at enterprises looking to adopt a high-touch, OpenAI-driven deployment process. Whether it evolves into a widely available platform or remains a closely managed product for select customers will depend in part on the answers OpenAI has yet to provide.
