Former OpenAI safety leader says firm warned of AI risks while deploying

By Splitfeed EditorialLast updated Oct 8, 2026, 16:50 UTC

On October 8, 2026, a former OpenAI safety leader publicly alleged that the company raised internal warnings about AI risks even as it continued to train and deploy AI systems. The claim centers on Sam Altman's OpenAI, and Zico Kolter, associated with OpenAI's safety work, was separately referenced in reporting from the same day. The specific systems, the timeline of those warnings, and the exact nature of what was flagged internally are not detailed in the available reporting.

The allegation matters because it fits a pattern of scrutiny OpenAI has faced over whether its public commitments to safety align with its operational decisions. If accurate, it would suggest a gap between what the organisation knew or feared and what it chose to do, which is precisely the kind of institutional behavior that regulators, researchers and the public have been pressing the company on. What remains genuinely in dispute is whether raising safety warnings while continuing development represents ordinary risk management, common across any technology company, or something more troubling, a deliberate choice to proceed despite serious internal doubt.

None of the three camps had published reactions by the time of this report, so what follows reflects what each would typically argue. The Pro-AI camp would likely contend that acknowledging risk while moving forward is responsible engineering practice, not hypocrisy, and that no serious technology is developed in the absence of uncertainty. The Anti-AI camp would almost certainly treat the allegation as confirmation that frontier AI labs prioritise competitive deployment over genuine safety, and would call for stronger external oversight or outright development pauses. The Middle Ground camp would probably argue that internal warnings are only meaningful if they produce binding constraints, and would push for transparent governance structures that make such concerns visible and actionable before deployment rather than after the fact.

The next significant development to watch is whether the former safety leader provides specific documentation, names the systems or dates involved, or whether current OpenAI figures including Kolter respond publicly. Any formal regulatory inquiry prompted by the allegation would also shift this from an internal dispute into a matter of legal or policy consequence.