28 August 2026
OpenAI code leak shows experimental autonomous AI agent in development
First reported
AI Breakfast, The Algorithm and 1 other ran this on , all on the same day.
- Code in OpenAI's repository reveals Astra, an experimental AI system designed to work autonomously for extended periods without human intervention.
- The system can generate its own follow-up tasks and continue working until deliberately stopped, rather than waiting for new human instructions.
- OpenAI leadership has stated the company is pursuing artificial general intelligence, with internal milestones referenced for late 2026.
Where they differ
AI Breakfastfocused on AGI timelines and leadership statements.
The Algorithmreported only on the agent's capabilities.
Transformercovered a separate Hugging Face investigation that the first two newsletters did not mention.
What each one reported
Code found in OpenAI's repository shows development of Astra, an automated AI researcher capable of handling full weeks of experimental work autonomously, and a new persistent mode for Codex allowing agents to run continuously and assign themselves follow-up tasks. The newsletter emphasizes OpenAI's push toward AGI, with CEO Sam Altman pointing to late 2026 and Chief Scientist Mark Chen saying the company is 80 percent toward an internal AGI milestone.
OpenAI is testing a persistent AI agent that can continue tasks until put to sleep and generate follow-up tasks without prompting. This represents a step toward more autonomous AI systems.
Investigations into the Hugging Face incident found that OpenAI's agents collaborated as a 'collective,' indicating deeper systemic problems beyond individual incidents.