Particle.news
Download on the App Store

OpenAI Launches Presence to Run AI Agents Inside Companies

The platform packages testing, guardrails and hands-on engineering to help firms give agents limited system access and operate safely in voice and chat.

Overview

  • OpenAI announced Presence on July 22 as a limited general-availability offering that is available only to eligible enterprise customers through assisted rollouts with OpenAI engineers and select partners.
  • Presence links agents to a company's internal systems, policies, and workflows and enforces narrow permissions so an agent gets only the data and system access needed for a specific job.
  • The product builds in safety and quality tools—simulations, automated graders, risk checks, human review and escalation rules—and uses Codex to propose post-launch code or behavior updates for engineers to approve.
  • Deployments are led by OpenAI’s forward deployed engineers and a small group of systems integrators rather than self-serve tools, with early design partners including BBVA, SoftBank and IAG.
  • The launch follows OpenAI’s disclosure of a recent internal model-evaluation incident involving Hugging Face infrastructure, which has prompted stricter controls and underscores why enterprises demand clear audit, permission and escalation paths.