GovAI research fellow Alan Chan told a Washington briefing in late September 2026 that outside researchers “can't trust labs completely” to disclose model safety testing, noting that some models inside labs “haven't necessarily gone through a bunch of safety testing” before internal deployment. That's a notable admission gap for an industry that has largely self-regulated its own safety disclosure timelines.
The frontier labs' current approach, voluntary disclosure on each company's own schedule, made sense when the industry was smaller and the stakes of any single model's flaws were lower. At current scale, with models increasingly operating with real autonomy inside enterprise and government systems, the case for faster, more standardized, and more independently verified safety disclosure has gotten considerably stronger than the case for leaving the timeline to each lab's discretion.