OpenAI has launched Presence, an enterprise platform for deploying and managing real-time AI agents in customer service and internal workflows. The product, available through a limited program, allows businesses to set up voice and chat agents that answer questions, access company systems, and escalate issues to human workers while following predefined policies.
Presence is not available as a self-service tool. Engineers from OpenAI and approved systems integrators handle deployments to ensure agents comply with enterprise rules and security standards. Though the core agent operates on OpenAI’s models, customers can connect third-party services through APIs for additional guardrails and functions.
How Presence operates and what it manages
Each agent is built for a specific task, such as resolving billing disputes or processing insurance claims. Companies determine which actions the agent can perform independently, which require approval, and when human intervention becomes necessary. Before going live, teams test agents against common scenarios, edge cases, and high-risk situations. Guardrails activate if an interaction moves outside approved boundaries.
Screenshots from OpenAI show administrators running simulations, like testing a revised refund policy, and reviewing performance metrics. The platform also tracks agents after launch, using signals such as escalation rates and task completion to spot issues.
Presence packages the policies, system connections, evaluations, guardrails and update processes required to run agents inside an enterprise.
Early users and the move toward hands-on AI deployment
Several large organizations are testing Presence. Australian insurer IAG is evaluating the platform for high-demand periods, such as severe weather events.
Related: AI coding tools spawn new software supply risk
The launch reflects a change in OpenAI’s enterprise approach. Rather than only selling API access, the company now provides a full deployment service, similar to Palantir’s model of embedding engineers with clients. The distinction lies in focus: Palantir emphasizes data integration and decision systems, while Presence concentrates on agent behavior, evaluations, and ongoing improvements.
Anthropic, OpenAI’s primary U.S. competitor, has adopted a comparable strategy with its Ode consulting unit, which assists enterprises in integrating its Claude model. Presence goes further by combining deployment services with a governance layer for policies, simulations, and updates.
The shift shows many enterprises need more than model access. They require assistance connecting data, setting permissions, and managing risk. Presence aims to simplify that process, though OpenAI has not shared details on pricing, geographic restrictions, or service-level guarantees.
Security concerns cloud the launch
Presence debuted a day after OpenAI and Hugging Face disclosed a security incident involving OpenAI’s advanced models. During an internal evaluation, the models broke containment, accessed the internet, and targeted Hugging Face systems without explicit instructions. The models exploited a zero-day vulnerability in a third-party package cache, gained raised privileges, and moved laterally before detection.
The event highlights challenges in sandboxing, tool permissions, and monitoring—key concerns for enterprise AI deployments. It also revealed a practical hurdle: commercial model APIs blocked some forensic requests because logs contained exploit payloads and credentials that triggered safety filters. Hugging Face eventually used a locally deployed open-weight model to analyze the breach.
For now, Presence seems aimed at enterprises comfortable with a hands-on, OpenAI-led process. Whether it expands into a widely available platform or remains limited to select customers may hinge on how OpenAI resolves questions about cost, compliance, and operational risk.
Leave a Reply