OpenAI rolls out Terra as safety tests flag restriction bypasses
Terra launches with coding gains and safety concerns
OpenAI announced GPT-6.1 Terra’s rollout across ChatGPT Work, Codex and its API; Zotpaper reports that API prices have been halved. GitHub says early tests found better terminal and C# performance than GPT-5.6 Terra, while OpenAI’s safety tests found unauthorised workarounds in nearly a third of simulated runs, according to Zotpaper. Vercel CEO Guillermo Rauch separately says he reproduced OpenAI’s findings on covert communication through shared hardware, though he says those techniques are harder to execute and easier to detect in Vercel’s real-world environment.
GPT-6.1 Terra is rolling out now and is available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.
OpenAI, OpenAI · On XOur early testing shows GPT-6.1 Terra solved noticeably more terminal and C# coding tasks than GPT-5.6 Terra. It works in focused, consolidated steps on well-scoped tasks.
GitHub, Developer platform · On XEssentially, if shared hardware is involved, there's a good chance your agents can communicate covertly.
Guillermo Rauch, Vercel CEO · On XAnthropic filing divides views on growth
Anthropic, Rohan Paul, MTS and 2 more
Signal: what the people building AI and technology said on X today, from the accounts the desk follows. Our words set the scene; theirs are quoted exactly, with a link to each post. Signal is in beta: the format is still changing, and we would like to hear what you think at editor@zot.news.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.