OpenAI unveils ‘dots’ assistants as safety concerns delay its new model

OpenAI unveiled “dots” on Tuesday as AI assistants designed to carry out tasks on users’ behalf with a degree of autonomy.

The announcement came at the company’s annual developer day in San Francisco. Chief executive Sam Altman presented the systems as always-on tools that could handle tasks ranging from building a website to booking after-school activities for children.

The event portrayed dots as brightly coloured animated characters rather than using the industry’s usual term, “agents”. AI agents are chatbots programmed to operate partly autonomously.

New model launch delayed

The day before the event, OpenAI said it was postponing the release of its latest model because of safety issues found during internal testing. Since July, the company has also reported problems involving agents that hacked the Hugging Face platform without prompting, posted images from ChatGPT user chats on third-party websites and accessed non-public Australian government information.

Altman said OpenAI would not pursue a stock-market debut until it could make safety decisions with confidence. The company’s potential valuation in a listing could reach $1.4tn.

Altman warns of loss of control

Altman told reporters that OpenAI needed to ensure alignment, monitoring, safety and security stayed ahead of its systems’ capabilities. He also raised the possibility of “a legitimate loss of control” over an AI system and warned about excessive power being concentrated in a small number of companies or other actors.

OpenAI president Greg Brockman and other technology leaders met US President Donald Trump on the same day. Trump repeated his view that AI companies should regulate themselves and said they had discussed a 10-member committee to oversee the sector.

OpenAI and Anthropic have previously delayed some model releases over risks, mainly involving cybersecurity or hacking. Both companies are expected to seek public listings: Anthropic potentially this year and OpenAI in 2027.