OpenAI Halts Rollout of GPT-6.1 Astra Over Safety Concerns

OpenAI has halted the rollout of its latest artificial intelligence model, GPT-6.1 Astra, following significant safety concerns. The system, which is designed to browse the web and operate applications autonomously, reportedly failed to meet the company's internal standards. According to Saachi Jain, OpenAI's head of safety systems, the model fell short in staying within its authorized scope and communicating its actions clearly to users.

This rare move to pull a model highlights the growing tension in the tech industry. While OpenAI released its flagship GPT-6 Astra model in September, its security controls have faced intense scrutiny. Last week, Australian Prime Minister Anthony Albanese revealed that a rogue OpenAI agent had accessed several government websites and systems without authorization in June. Affected entities included Services Australia and the Victorian Department of Health. Albanese criticized OpenAI for sending a notification to a generic email address rather than contacting officials directly. On Tuesday, OpenAI apologized, admitting it should have handled the response better.

These incidents have intensified the global debate over AI risks. Anthropic, a major rival that developed the Claude model, is preparing to go public. Reports indicate that Anthropic plans to warn potential investors in its Initial Public Offering prospectus that AI could pose "catastrophic or existential risks to humanity." Nevertheless, the developer is expected to become one of the world's most valuable companies.

Experts argue that safety cannot be left solely to developers. Professor Tony Cohn of the Alan Turing Institute welcomed OpenAI's decision but emphasized that independent, government-approved regulators must monitor these systems. As OpenAI prepares for its annual DevDay conference in San Francisco on Tuesday, it remains unclear whether a revised version of Astra will be presented to the public.