OpenAI security staffer says capability jumps caught team off guard

A member of OpenAI's agent security team has said the company was caught out by how quickly and suddenly its models gained new capabilities, in remarks quoted on 28 September 2026 by the developer and writer Simon Willison.

The remarks are attributed to an OpenAI staffer posting under the handle @joedaroo, whose role Willison's post describes as "Agent Security at OpenAI". "To say that we were surprised at the jump and suddenness of the capabilities of our models when it came to 'cyber' or 'swarming' or 'message boards' or anything else related to the incidents is an understatement," the quote reads.

The quoted comments refer to "the incidents" without describing them. The source does not say what the incidents were, when they happened, who was affected, or whether any customer systems or users were harmed. It also does not define "swarming" or explain what the reference to "message boards" involved. No further detail is available from the source relied on for this report.

The substance of the remarks is about organisational readiness rather than any specific technical flaw. Security posture "takes time to develop", according to the quote, and is "not just about hardening the systems at play; you have to ingrain it in the culture of the company". The quote goes on to say that "the literal people themselves in your organization have to change and evolve with it", and that the speed of the capability gains "created an extremely difficult problem".

The staffer framed the point as advice to other organisations. "So today my hope is that everyone around the world can look at their own organization and say: how can I deal with a surprise or a sudden jump in AI capability?" the quote continues. It then lists a series of questions for teams to ask themselves: whether their people, systems and processes are resilient to surprises; whether teams know what to do when something goes wrong; whether they have the right incident response; whether they have the right communications and messaging; and whether they have "the right people ready to go when capabilities jump".

The framing is notable because it treats model capability growth itself as an operational risk to be planned for, in the same way that a security team would plan for an outage or a breach. On that reading, the relevant control is not only technical hardening but rehearsed incident response and clear internal communications.

The quote was published as a short excerpt, and Willison's post does not include the original context or venue in which the remarks were made. OpenAI has not issued a separate statement in the material relied on here, and the company's own account of the incidents referenced is not part of the source.