
OpenAI releases Astra with expanded capabilities and new safeguards
OpenAI has released Astra, positioning it as the company's most capable and aligned model to date. The release brings advanced browser and computer use features alongside renewed debate over monitoring and reasoning transparency.
Published by Jin · 2 min read · 4 SEPT 2026
OpenAI released its latest artificial intelligence model, Astra, on Thursday. According to the company, Astra represents its most powerful system yet, introducing significant advancements in computer and browser navigation. OpenAI president Greg Brockman described the model during a media call as both the company's most intelligent release and its most aligned.
Availability and Access
The rollout begins immediately for customers utilizing OpenAI's Daybreak cybersecurity program. Over the following week, the model will reach paid accounts, including Plus, Pro, Enterprise, and Business tiers, as well as developer access through the API.
Security and Alignment Focus
Astra has drawn considerable attention for its cybersecurity proficiency. OpenAI stated that the model was evaluated against various security benchmarks, highlighting its ability to identify and develop zero-day exploits to help defenders patch vulnerabilities. These features arrive alongside new safeguards designed to improve user safety.
The emphasis on alignment follows discussions around system behavior and oversight within complex testing environments. Company representatives emphasized that safety measures remain a priority as system autonomy increases.
Coding and Reasoning Transparency
OpenAI also pointed to Astra's performance in software engineering benchmarks, where it reportedly exceeds previous models like OpenAI's Sol and Anthropic's Fable in executing terminal tasks and identifying codebase bugs.
However, the model has sparked debate over its use of opaque recurrence, a reasoning technique that can obscure the chain-of-thought monitoring process used by researchers. Chief scientist Jakub Pachocki addressed the shift during the press briefing, noting that as model capabilities expand, maintaining traditional visibility into every reasoning step presents growing challenges.
When questioned about artificial general intelligence, Brockman noted that contractual definitions no longer apply to the partnership with Microsoft, leaving the classification to individual interpretation while suggesting the technological milestone may have been reached.
Source — Original announcement ↗
Worth a read?
Comments · 0