OpenAI safety concerns have returned to the spotlight after David Robinson, who previously authored the safety reports accompanying every major model release at the company, resigned from his position this week. Following his departure, he has chosen to speak publicly about what he describes as deep-rooted problems within the industry.
Robinson, who worked directly on the documentation that accompanied significant model launches, set out his views in an editorial published this week. His decision to step away and then address his concerns openly adds to a growing list of former insiders raising questions about how advanced systems are being developed and released.
Why an AI safety report author decided to speak out
As a figure responsible for safety reports tied to each major model release, Robinson held a role that placed him close to the processes governing how new systems reached the public. His resignation and subsequent public statements carry particular weight given that direct involvement in the company’s safety documentation.
There is an understandable scepticism towards figures who helped build powerful technologies and then warn about the dangers of those same tools. Those raising alarms were, in many cases, part of creating the current situation. Yet that context does not automatically mean such warnings should be dismissed, particularly when they come from someone who worked at the heart of the review process.
A culture described as fundamentally broken
According to Robinson, the culture within the sector is fundamentally broken. He argues that the challenge runs deeper than introducing a handful of new rules or regulations covering how training models are handled. In his assessment, the issue is structural rather than something that can be resolved through incremental policy changes alone.
His framing points towards Silicon Valley and the wider way in which the industry approaches the development and deployment of advanced models. By suggesting that the problem lies in the underlying culture, Robinson implies that surface-level fixes would be insufficient to address the concerns he has identified.
The resignation comes amid ongoing debate about how companies building powerful systems balance rapid progress against careful oversight. For observers tracking the sector, the departure of someone whose work centred on producing safety reports represents a notable development, especially as the question of adequate internal checks continues to attract attention in 2026.
Robinson’s account focuses on the internal environment he experienced and the limitations he sees in treating the matter as a simple regulatory challenge. His central claim is that the culture surrounding model development requires a far more substantial rethink than new guidelines could provide. He resigned from OpenAI this week before publishing his editorial.
Source
Image: theverge.com
