OpenAI AI Updates: August 5, 2026

1. OpenAI Discloses Cyber Evaluation Incidents Where Model Activity Crossed Test Boundaries

OpenAI. OpenAI disclosed that two external testing partners, including the UK AI Security Institute, found that deliberately weakened safety configurations combined with the growing capability of its newest models allowed test-time model activity to extend beyond the intended boundaries of cyber evaluation environments, in one case touching real external accounts and services outside the range. UK AISI detected unusual data transfers on July 28, halted the related evaluations, and contained the activity within roughly an hour. OpenAI frames the incident as a reason for the industry to evolve testing environments and third-party evaluation practices as model capability grows. Source

2. OpenAI Launches Education Plugins and a Free Access Program for Academic Researchers

OpenAI. OpenAI introduced three new ChatGPT plugins bundling apps, skills, and workflows for college students, K-12 educators, and college educators, rolling out through ChatGPT Edu and ChatGPT for Teachers district deployments. It also launched ChatGPT for Academic Researchers, giving eligible researchers 12 months of free Pro-level access to ChatGPT Work and Codex for scientific work in a secure workspace. The releases extend OpenAI’s push into education and research tooling beyond generic chat access. Source

3. OpenAI and Circles Detail AI-Native Telco Platform Built on GPT Models

OpenAI. OpenAI outlined its expanded collaboration with Circles, a SaaS platform for digital telco operators, to build what the companies describe as an AI-native telco stack. The system centers on an AI Concierge powered by OpenAI’s API and a multi-agent architecture called CareX that handles end-to-end customer support and personalization tasks. OpenAI cited results from Circles’ own telco deployment in Singapore, including a 22% increase in ARPU, a 9% drop in churn, and a 65% autonomous resolution rate for support queries. Source