Ex-OpenAI Policy Chief Brundage Criticizes Shipping With Known Flaws and Industry Safety Gap

Miles Brundage, former OpenAI head of Policy/AGI Readiness, posted a series of statements on X from October 3–4 about "iterative deployment" and industry safety. His core argument: criticizing the framing of "forced rapid releases" as a safety virtue, and pointing out that the industry's overall safety level falls short. After sparking debate, he made several clarifications and additions.

Confirmed

• Brundage acknowledged that learning from real-world use after deployment is legitimate—some lessons only come from shipping. His criticism targets releasing products known to be defective, which doesn't accelerate learning, only shipping.
• Amid the debate, he clarified he wasn't assuming bad faith, but rather lacked patience for point-by-point responses "with only hundreds of thousands of minutes until superintelligence." His core point: companies and insiders have incentives to repackage necessity as virtue.
• On liability for deaths, he explained his earlier wording was deliberately conservative, using "tied to deaths" rather than claiming companies bear full causal responsibility; but he stressed that at a scale of billions of users, if one believes the latter has never happened even once, he has nothing to say.
• On industry safety, he made two arguments: this is an industry-wide problem, not unique to OpenAI; and even if you combined the best safety practices and safety culture of every company into one, it would still fall below his acceptable standard (the post's analogy referencing "below Chernobyl..." was truncated).

Why it matters

• Brundage served as OpenAI's head of Policy, head of AGI Readiness, and in senior safety roles—his insider critique offers a reference point for evaluating frontier labs' safety practices.
• He distinguishes the industry's familiar "iterative deployment" defense from "shipping with known defects," exposing how that rhetoric may paper over safety issues.
• His statement that lethal liability cannot be zero for products with billions of users takes a clear stance in the debate over AI companies' boundaries of responsibility.

2026-10-03 ~ 2026-10-04 · 5 related posts

• [source] Miles Brundage: shipping with known flaws isn't iterative deployment, it's just faster shipping — Miles_Brundage · 2026-10-03
• Brundage doubles down: with billions of users, AI firms aren't causally blameless for deaths — Miles_Brundage · 2026-10-03
• [source] Ex-OpenAI risk lead Brundage: an industry with billions of users has zero AI-death cases? You're not thinking clearly — Miles_Brundage · 2026-10-04
• [source] Ex-OpenAI researcher: best practices from all AI labs combined would still be sub-Chernobyl — Miles_Brundage · 2026-10-04
• Miles Brundage: 'iterative deployment' is often expediency dressed up as safety — Miles_Brundage · 2026-10-04