The Second Fold

When a frontier lab ships faster than its safety team can keep up, safety stops being an independent function and becomes a sub-team of the faster org. OpenAI has now run that play twice in 24 months, and the second time the receiving research lead also holds 'Safety' in her title. Call it the second fold.

The first fold dissolved the Superalignment team in 2024 after Leike and Sutskever left, with its promised 20% of compute quietly reduced. The second fold puts Safety Systems inside Mia Glaese's research vice presidency, with Saachi Jain reporting into her as interim lead. The org chart no longer has a safety line that does not run through a research boss.

Mark Chen's framing of the Heidecke exit makes the mechanism explicit. Chen told WIRED that compressed training and release cycles now produce safety coordination challenges larger than any prior period. Read that line carefully: faster ship cadence is the cause; safety's loss of independence is the effect. The faster the model release clock, the shorter the leash — in the reporter's reading of the pattern — on the team tasked with questioning the next release.

Read it as body-count — six safety leads gone in two years — and the story is morale. Read the chart, and the story is that the safety function no longer has an independent voice in the room. The license to operate is the reporting line, not the personnel file. Watch the next safety lead's title: if 'safety' sits under a research VP, the second fold has become the standing structure.