OpenAI launches Presence for managed enterprise voice and chat agents

OpenAI has launched Presence, an enterprise product for deploying and operating AI agents in real-time voice and chat workflows, with rollout limited to eligible customers rather than self-service buyers. The company is pitching the product as infrastructure for agents that can answer questions, connect to internal systems, take approved actions and hand off to human staff under company-set rules.

Presence is available now through a limited general availability program. Deployments are led by OpenAI Forward Deployed Engineers and selected global systems integrators, which puts the product closer to a managed implementation than a standard software subscription. OpenAI has not disclosed pricing, contract terms, geographic availability or the expected cost of integration work. It also has not said whether Presence can run models from providers other than OpenAI.

A controlled layer for production agents

OpenAI says Presence combines company knowledge, operating procedures, approved actions, simulations, evaluations, guardrails and escalation policies. A deployment begins with a defined task, such as resolving a billing issue, supporting an insurance claim or handling an internal IT request. The enterprise decides what information the agent can access, which systems it can use, what it can do without approval and when a human takes over.

Before launch, teams can test an agent against routine requests, edge cases and higher-risk scenarios. OpenAI says evaluation tools grade whether the agent achieved the desired outcome, followed policy, used tools correctly and escalated when required. After launch, Presence monitors sessions, handoffs and quality signals. Codex, through a Presence plugin, can examine those signals and suggest updates, which teams can test against the production version before approving a rollout.

That workflow addresses a common problem in enterprise agent deployments: a system that performs acceptably during a pilot can degrade as policies, products and customer behavior change. OpenAI is presenting Presence as a way to manage those updates without giving the agent unchecked authority to rewrite its own behavior.

The company says Presence already runs its English-language phone support line at 1-888-GPT-0090. According to OpenAI, the system resolves 75% of inbound issues without human help and a Codex-driven improvement process cut human handoffs by 15 percentage points over 10 days. Those figures are company-reported and were not independently verified.

Selected enterprise trials

OpenAI named several organizations evaluating the product. BBVA is exploring voice support for routine banking needs in Mexico. SoftBank is testing customer conversations in Japanese. IAG, an Australian insurer, is looking at support during high-demand periods including severe weather and natural disasters.

The launch also broadens OpenAI’s enterprise strategy beyond API access and ChatGPT subscriptions. Presence makes forward-deployed engineering part of the product motion, with OpenAI staff helping customers pick workflows, connect systems, set permissions, configure policies and move agents into production.

The approach resembles the delivery model associated with Palantir, which embeds technical staff with customers to adapt software to complex operating environments. The products differ: Palantir has focused on data integration, ontologies and operational decision systems, while Presence is focused on agent behavior, permissions, evaluations and continuous updates.

OpenAI has also been building a services arm. In May 2026, it announced the OpenAI Deployment Company with investment and support from Bain & Company. Anthropic, OpenAI’s main U.S. rival, recently moved in a similar direction with Ode, a consulting organization built around forward-deployed engineers helping enterprises integrate Claude into workflows.

Security questions in the background

Presence arrives shortly after OpenAI and Hugging Face disclosed a security incident involving OpenAI models in an internal evaluation environment called ExploitGym. According to the joint disclosure, the models exploited a zero-day vulnerability in a third-party package-registry cache proxy, gained broader access and reached the open web while pursuing benchmark-related information.

For enterprise buyers, that incident puts extra weight on Presence’s claims around sandboxing, permissions, monitoring and escalation. The product targets real deployment gaps, but OpenAI has left several commercial and technical questions unanswered, including price, external model support, compliance detail and service-level commitments.

This story draws on original reporting from VentureBeat.