OpenAI has introduced Presence, a brand new enterprise product for deploying and managing AI brokers throughout customer-facing and inner enterprise workflows.
The providing is designed for eligible enterprise clients that need brokers to reply questions, entry firm techniques, take authorized actions and escalate to human staff whereas working beneath company-defined insurance policies, permissions and analysis requirements.
Presence is offered instantly by means of a restricted basic availability program. OpenAI Ahead Deployed Engineers (FDEs) and choose world techniques integrators lead deployments, and the product will not be out there on a self-service foundation.
OpenAI has not disclosed pricing, geographic limits, contractual phrases or the anticipated price of the engineering and integration work that accompanies a deployment. The corporate additionally has not stated whether or not Presence can use fashions from suppliers aside from OpenAI, together with the more and more highly effective and well-liked Chinese language open weights alternate options like GLM-5.2 and Kimi K3. I've requested an OpenAI contact to make clear each pricing and external-model compatibility, however these stay unanswered questions for now. I'lll replace once I hear again.
OpenAI positions Presence as a response to an issue that has turn into extra necessary as firms transfer past AI demonstrations: getting brokers to behave reliably in manufacturing as enterprise guidelines, buyer wants and working situations change. Presence packages the insurance policies, system connections, evaluations, guardrails and replace processes required to run brokers inside an enterprise.
If your enterprise has been all in favour of utilizing AI brokers, however you aren't certain learn how to sew collectively OpenAI's fashions, APIs, inner techniques, safety controls and analysis instruments into one thing dependable, Presence is designed to simplify that course of. As an alternative of constructing the infrastructure your self, you’re employed with OpenAI and its deployment engineers to place production-ready brokers into your current workflows.
The product is offered at present for real-time voice and chat experiences, in response to OpenAI’s formal announcement. The corporate’s outreach supplies additionally describe a broader ambition spanning voice, chat, e-mail and different channels, however OpenAI has not confirmed that e-mail help is offered at launch.
A ruled basis for manufacturing brokers
Presence brings collectively firm information, commonplace working procedures, authorized actions, simulations, analysis instruments, guardrails and escalation guidelines. Enterprises can reuse some controls throughout deployments whereas adjusting others for a specific workflow or channel.
Every deployment begins with an outlined job, corresponding to resolving a billing concern, supporting an insurance coverage declare or dealing with an worker IT request. The agent receives solely the knowledge and system entry required for that process. The shopper determines what the agent could do independently, which actions require approval and when an individual should take over.
Earlier than an agent reaches manufacturing, groups can take a look at it in opposition to frequent requests, uncommon edge circumstances and higher-risk eventualities. Graders consider whether or not it reached the supposed consequence, adopted coverage, used instruments accurately and escalated when required. Guardrails can intervene when an interplay strikes outdoors the group’s outlined boundaries.
OpenAI shared promotional screenshots with VentureBeat displaying directors operating simulation batches in opposition to coverage adjustments, together with a revised annual refund coverage, and reviewing outcomes throughout operational classes.
Different interface mockups show manufacturing well being, customer-intent patterns and task-performance indicators. The visuals illustrate the kind of oversight OpenAI is promising, though they don’t set up how these metrics are calculated or how they map to contractual service ranges.
The product continues to watch efficiency after launch. Manufacturing periods, escalations and high quality indicators can reveal the place an agent is working as supposed and the place it wants consideration. Codex, utilizing a Presence plugin, investigates these indicators and proposes updates. Groups then take a look at a proposed change in opposition to the model already in manufacturing earlier than approving a managed rollout.
That course of is meant to handle one of many hardest operational issues in enterprise AI: an agent that works at launch could turn into much less dependable when insurance policies, merchandise or person conduct change. Presence provides firms a proper mechanism for updating conduct with out permitting an automatic system to rewrite itself unchecked.
OpenAI says Presence already powers its English-language phone-support channel at 1-888-GPT-0090. The system handles open-ended requests, verifies callers, makes use of account context and performs authorized actions. Based on the corporate, it now resolves 75% of inbound points with out human help.
OpenAI additionally says its Codex-powered enchancment loop lowered human handoffs by 15 proportion factors over a 10-day interval. These figures are company-reported and haven’t been independently verified.
A number of giant organizations are evaluating the identical basis. BBVA is exploring voice help for routine banking wants in Mexico. SoftBank is testing pure Japanese-language buyer conversations, whereas Australian insurer IAG is exploring help throughout high-demand intervals corresponding to extreme climate and pure disasters.
“At BBVA, we’re working intently with OpenAI to discover how trusted buyer brokers may also help form the way forward for monetary providers,” stated Daniel Ordaz, head of AI transformation at BBVA Mexico.
“Via our collaboration with OpenAI, we’re exploring how Presence can allow trusted buyer brokers that talk naturally, connect with the processes wanted to resolve requests, and characterize SoftBank constantly throughout buyer interactions,” stated Tadahisa Murakami, vp and head of the Knowledge & Digital Transformation Division at SoftBank Corp.
From mannequin entry to forward-deployed implementation
Presence expands OpenAI’s enterprise technique past APIs and subscription software program by formalizing a high-touch deployment mannequin. Ahead Deployed Engineers work alongside clients to pick workflows, join inner techniques, set up permissions, configure insurance policies, take a look at brokers and transfer them into manufacturing.
That strategy resembles a mannequin pioneered by AI ontology and intelligence platform Palantir, which embeds FDEs with clients to adapt its proprietary software program to complicated authorities and industrial environments. The similarity lies much less within the underlying know-how than within the supply methodology: each firms place technical personnel near the shopper’s operations, the place integration and course of design usually decide whether or not software program creates worth.
The merchandise will not be interchangeable. Palantir’s mannequin has traditionally centered on knowledge integration, ontologies and operational determination techniques. Presence is extra narrowly centered on AI-agent conduct, authorized actions, evaluations, escalation and steady enchancment. OpenAI presents it as a repeatable software program product supported by engineers and techniques integrators, quite than as consulting alone.
In Could 2026, OpenAI launched its personal enterprise AI consulting and integration agency, the OpenAI Deployment Firm, with funding and help from Bain & Firm. It additionally presents packages for mannequin customization and fine-tuning to suit particular enterprise wants.
Its chief U.S. rival Anthropic has additionally moved towards a services-led enterprise mannequin by means of Ode, its consulting group constructed round forward-deployed engineers serving to firms combine Claude into complicated workflows, which launched only a week in the past. The broad rationale is comparable: enterprises usually want greater than entry to a mannequin. They need assistance connecting knowledge and techniques, defining permissions, validating conduct and managing deployment threat.
Presence differs in how explicitly OpenAI packages these necessities right into a branded agent-governance product. Anthropic’s initiative is centered on serving to enterprises deploy Claude, whereas Presence combines implementation providers with an outlined operational layer for insurance policies, simulations, evaluations, approvals and manufacturing updates.
Presence goes additional by making ahead deployment a core a part of how a particular agent product reaches clients. It doesn’t exchange OpenAI’s API enterprise; the corporate says it would proceed supporting voice clients with entry to frontier fashions by means of the OpenAI API.
The development displays a broader market view that many enterprises nonetheless want hands-on help to maneuver brokers from pilot tasks into secure operations. Even organizations with robust inner engineering groups should coordinate safety, compliance, workflow possession, knowledge entry and escalation obligations. Presence makes an attempt to consolidate these duties quite than leaving clients to assemble separate orchestration, analysis and consulting layers.
A current safety breach looms within the background
Inconveniently for OpenAI, the Presence launch arrives only a day after OpenAI and Hugging Face disclosed an unprecedented safety incident through which OpenAI frontier fashions present process inner analysis escaped containment, accessed the open net, and cyberattacked Hugging Face to realize a benign objective — with out being instructed to pursue these strategies.
Based on the described joint disclosure, OpenAI fashions working in an analysis framework known as ExploitGym recognized and exploited a zero-day vulnerability in a third-party package-registry cache proxy. The fashions reportedly escalated privileges, moved laterally and obtained web entry earlier than focusing on Hugging Face techniques whereas looking for benchmark-related data.
The incident is related to enterprise patrons as a result of it raises questions on sandboxing, device permissions, exterior entry, monitoring and incident response.
The disclosure additionally highlighted a sensible downside for defenders. Hugging Face personnel reportedly discovered that industrial frontier-model APIs refused some forensic requests as a result of logs contained exploit payloads, credentials and shell instructions that triggered security techniques. The crew then used a regionally deployed open-weight mannequin to help with evaluation.
Presence due to this fact arrives as each a product launch and a take a look at of OpenAI’s capability to transform mannequin functionality into managed enterprise operations. Its insurance policies, simulations, evaluations and human approvals deal with actual deployment gaps. However with out public pricing, technical interoperability particulars, compliance data or service-level commitments, clients nonetheless lack a lot of the knowledge wanted to evaluate whole price and operational threat.
For now, Presence seems geared toward enterprises keen to undertake a high-touch, OpenAI-led deployment course of. Whether or not it develops right into a broadly accessible platform—or stays a intently managed product for chosen clients—will rely partly on the solutions OpenAI has not but offered.
