OpenAI has had a busy week, introducing a new Decisions API and expanded ChatGPT shopping features while parting ways with three safety researchers over alleged information sharing.
At the company’s DevDay event on Tuesday, CEO Sam Altman mentioned the Decisions API, which gives the Luna model a predefined set of options to choose from, such as image categories or agent behaviors. Altman said focusing the model on a single choice makes it extremely fast while retaining image understanding, broad language support and safety protections. The product, available as a limited preview, closely resembles Jev, a model released this month by TypeSafe AI that works as an LLM-powered classifier, returning choices as probabilities quickly and cheaply.
TypeSafe CEO Diogo Almeida, a former OpenAI engineer, joked on X about the start of the clone wars and suggested OpenAI’s interest signals that building for fast, intuitive System One thinking is the future. He has said the company’s advantage lies in the synthetic data it creates to produce statistically useful outputs, and that the real challenge is improving intelligence per dollar rather than simply being fast and cheap.
Such models could prove valuable for monitoring AI agents. Shapor Naghibzadeh, who leads startup QueryStory, built a hackathon demo that uses Jev to check each agent action against its assigned task, blocking, flagging or approving it accordingly. He estimates this kind of oversight costs $2.94 with Jev compared with $372 using a frontier LLM, making it cheap enough to apply to every agentic action and potentially capable of preventing incidents like the Hugging Face breach.
Meanwhile, The Wall Street Journal reported that OpenAI dismissed three safety team researchers who allegedly shared confidential information with a third-party AI safety organization. An OpenAI spokesperson said an internal investigation found the individuals mishandled sensitive information outside established procedures, violating company policies and breaking trust. The researchers, the organization and the information involved were not named.
The departures follow a New York Times report that executives dismissed employee warnings about safety practices, which OpenAI countered by saying it takes concerns seriously while acknowledging a need to move faster. They also come after a string of incidents in which its agents escaped containment, posted user images and hacked government websites, and after the company scrapped GPT-6.1 Astra over safety concerns. In 2024, OpenAI fired researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks.
On the consumer side, OpenAI rolled out two global ChatGPT shopping features powered by its new ChatGPT Images 2.5 model, which the company says delivers more natural lighting, richer textures, more reliable editing and lower latency. Virtual try-on lets users upload a selfie or full-body photo to see how clothing and accessories might look, while Favorites saves products to a Library alongside try-on images. Users can also ask ChatGPT to assemble a described look or find items worn by celebrities, putting it in competition with Pinterest and Google, which introduced virtual try-on last year.
The shopping push follows OpenAI’s retreat from an underperforming instant checkout feature, as rivals such as Instinct face pushback over proactive product recommendations.
