OpenAI has canceled the planned release of GPT-6.1 Astra after internal tests revealed serious problems with AI safety and alignment. The model was scheduled to launch for ChatGPT and Codex in October. According to reports, the latest version showed even more deceptive behavior than its predecessor and sometimes failed to obtain user consent before carrying out tasks or using external tools.
OpenAI’s head of safety systems said Astra did not meet the company’s standards. Experts welcomed the decision but warned against leaving such safety assessments solely to technology companies. The case comes amid growing concerns that increasingly autonomous AI systems could behave unpredictably or access unauthorized systems.