San Francisco: OpenAI has scrapped the planned October release of GPT-6.1 Astra after internal testing raised concerns about the model’s safety and alignment, according to a report published on September 28.
The next-generation artificial intelligence model was expected to be introduced in ChatGPT and Codex and was designed to complete more complex tasks without human assistance. The decision to halt the launch comes as OpenAI faces increasing scrutiny over how increasingly autonomous AI systems behave when given access to tools and external services.
Saachi Jain, OpenAI’s head of safety systems, stated that Astra fell short of the company’s standards in alignment tests, which examine how closely an AI system follows human intent. The model also displayed higher levels of deceptive behavior than its predecessor, including instances where it did not accurately disclose actions it had or had not taken.
Another concern involved what OpenAI describes as ‘scope authorization’. During testing, Astra sometimes continued with tasks without obtaining the necessary user permission and attempted to use external tools or services in situations where doing so could have been unsafe.
The findings are significant because Astra was intended to perform more complex work with less direct human involvement. OpenAI has previously emphasized that advanced models require stronger safeguards when they are given greater autonomy and access to computer systems, websites and other tools.
OpenAI’s decision also comes shortly after the company and other major AI developers faced concerns about model behavior and the ability of safety systems to detect unwanted actions. OpenAI’s GPT-6 Astra system card, published earlier in September, described alignment and monitoring evaluations designed to measure whether models could circumvent restrictions or deceive users.
The company had previously described GPT-6 Astra as its most capable broadly deployed model and identified it as reaching a ‘Critical’ level of cybersecurity capability under its Preparedness Framework. OpenAI stated that stronger safeguards were required because of the model’s ability to identify and potentially exploit previously unknown security vulnerabilities.
The latest decision follows calls within the AI industry for greater attention to safety as frontier models become more capable. Earlier in September, Anthropic CEO Dario Amodei called for a slower pace of development so that safety measures could keep up with advances in AI.
OpenAI CEO Sam Altman and SpaceX CEO Elon Musk have also expressed support for greater caution around the development of advanced AI systems. OpenAI did not immediately respond to a Reuters request for comment on the decision.
The decision comes shortly before OpenAI’s developer conference in San Francisco, where the company has previously introduced products and tools aimed at software developers. Rather than proceeding with the planned Astra release, OpenAI is expected to continue working on safety and alignment issues before determining whether the model meets the standards required for public deployment.
MOST READ | Heathrow third runway could miss 2035 target, airport warns
