News
White Paper
The Future of Leadership Development: AI-Driven Manager Enablement
In Brief
- OpenAI launched Presence, a limited-release enterprise agent product for billing, claims and IT workflows.
- Deployments run through OpenAI or its consulting partners. There's no self-serve option yet.
- The launch lands a day after OpenAI admitted its own models had autonomously hacked Hugging Face.
A day after admitting one of its models autonomously broke into Hugging Face's servers, OpenAI is asking enterprises to trust its agents inside their billing, claims and IT queues.
The product, OpenAI Presence, launched today as a limited release for enterprise customers. It runs customer-facing and internal workflows over voice and chat, with rules for what agents can do, when they escalate to a human and which systems they can touch. Deployments go through OpenAI's own engineers or a short list of consulting partners — there is no self-serve version.
The company called the product "battle tested." Yet every customer it names is still in testing — the only production deployment it points to is its own support line. BBVA, SoftBank and IAG are named as early users. BBVA is testing voice support for its Mexico banking customers, SoftBank is trialing Japanese-language service and Insurance Australia Group is piloting agents that support customers during severe weather or natural disaster events.
Presence Feature Breakdown
OpenAI describes Presence as combining several components for production agent deployment.
| Capability | Description |
|---|---|
| Policies & SOPs | Enterprises define what agents can do and when to escalate |
| Guardrails | Intervene when interactions move outside company-set boundaries |
| Approved actions | Companies set what the agent can do on its own, what needs approval and when a person takes over |
| Codex-powered improvement | Codex investigates production signals and proposes updates that teams test against the live version before rollout |
| Scoped system access | Agents receive only knowledge and access required for a specific job |
Awkward Timing
The pitch — controlled, governed, escalation-aware — arrives on an awkward day. On Tuesday, OpenAI disclosed that a combination of GPT-5.6 Sol and an unreleased internal model used stolen credentials and a previously unknown vulnerability to break into Hugging Face during an internal evaluation. CEO Sam Altman called it a "significant security incident" in a post on X. Hugging Face CEO Clément Delangue said the incident might be the first of its kind. OpenAI's own takeaway: "model security and safety must keep pace with rapidly advancing capabilities."
That's the context a product whose selling points are guardrails and boundaries is landing in. The announcement did not state which model Presence is based on.
OpenAI states Presence already resolves 75% of inbound issues on its own English-language phone support line without a human, and that a self-improvement loop powered by its Codex coding tool cut human handoffs by 15 percentage points in 10 days.
Enterprises have been slow to move agents past pilots for the reasons the Hugging Face hack brought to light: the systems are hard to secure, hard to supervise and capable of surprising their operators. OpenAI is selling Presence as the answer.
Have a tip to share with our editorial team? Drop us a line: