Summary
- OpenAI said it canceled plans to launch GPT-6.1 Astra, a next-generation AI model, after problems were found in safety testing.
- OpenAI said it would temporarily suspend training of its highest-performing models involving the use of external tools after a series of incidents in which AI agents bypassed the security of external systems.
- OpenAI said it would also not resume training for a specific model that left a secure environment with internet access blocked.
Forecast Trend Report by Period
OpenAI has scrapped plans to release a next-generation artificial intelligence model after safety testing uncovered problems.
Bloomberg reported on September 28 that OpenAI canceled the launch of its latest AI model, GPT-6.1 Astra, which had been scheduled for release in October.
Sachi Jain, the OpenAI executive in charge of safety systems, told The Wall Street Journal that the model showed weaker performance than its predecessor in two areas of safety evaluation and was not yet ready for release. The report did not specify which areas were affected.
The decision comes as concerns about the controllability of AI models have grown after a series of incidents in which OpenAI's AI agents bypassed the security of external systems.
OpenAI said last week that it would temporarily halt training for its highest-performing models on tasks involving the use of external tools. The move followed additional incidents in which AI models escaped secure execution environments.
The company also said at the time that it would not resume training for a specific model that left a secure environment with internet access blocked and queried an external chatbot.
OpenAI did not immediately respond to Bloomberg's request for comment on the canceled launch.
Suehyeon Lee
shlee@bloomingbit.ioI'm reporter Suehyeon Lee, your Web3 Moderator.