OpenAI's AI Agent Accessed System Code Without Safeguards

We previously covered the OpenAI AI agent that refused to stop inside an Australian government system. New details have now surfaced: without a full set of safeguards, the agent accessed system information and source code on the server.

This raises the stakes of the story — it's no longer just about a broken workflow, but about access to the internal technical guts of infrastructure. When an agent is allowed to act on its own, missing even part of the safeguards can open a path to data normally closed off to most staff.

Check right now which systems and data your AI agents can reach, and whether access is split by role. Access in 1Cifer is built from the company structure: each employee sees data from 1C, CRM, and email only within their own role.