OpenAI Slows Down Work on a New AI Model to Tighten Safety Standards
TL;DR
OpenAI is tightening its own safety measures for the training of its next model and is accepting a delay in development as a result. German outlet heise online reports this based on statements from the company. The timeline slips for a model that is already considered substantially more capable internally. For the market it signals that safety sign-off is becoming the pacing factor for release cycles.
Nauti's Take
The progress here is governance becoming concrete: a vendor that delays a release against its own safety criteria turns safety into a real planning constraint rather than a footnote. The limitation is that the criteria and the test results stay internal, so no outside party can check whether the delay was warranted.
Teams planning products around announced models should buffer their dates and keep the current generation viable as a fallback.
Sources