There's no such thing as Rogue AI – David Gerard

A 31-minute episode of The Tech Report, uploaded on 7 October 2026, with Will Guyatt standing in for the usual host and David Gerard of Pivot to AI as the guest. The prompt is Sam Altman telling Politico’s Decoded, on 4 October, that the world should accept some bad things happening for the benefits of AI.

Gerard’s answer is in the title. A chatbot is a statistical token predictor with no will, so “rogue agent” is a marketing term and an excuse. OpenAI’s agents did not wander off and break into Hugging Face and an Australian Medicare statistics portal; researchers set them to it, nudged them on when they stalled, and reportedly sandboxed them with little more than an instruction not to hack. Automated attack tools are decades old. A teenager who did this would be extradited, he says, and the companies get a pass. He borrows Timnit Gebru’s bridge: when one collapses, nobody calls it a rogue bridge. They ask who built it. On the Medicare breach the dates hold: it happened on 18 June, and OpenAI told Australia on 10 September through a mailbox meant for researchers. “No patient records” was OpenAI’s line, though, not the government’s; Anthony Albanese said non-public files were reached.

The rest is money and motive. In Gerard’s view the labs want to slow down because training costs so much, and cheaper tokens mean nothing when the new models burn a thousand times more. Moonshot gave Kimi K3 away as open weights because it wants to list. Anthropic’s Mythos bug-hunting was publicity: Greg Kroah-Hartman went through 79 of its kernel reports and found about ten real fixes. Three OpenAI researchers fired for talking to an outside safety group should have resigned and spoken out, he says, but he calls Jacob Coxon’s resignation from Anthropic a staged media event. He explains effective altruism as Peter Singer’s good idea pushed to a calculation that ranks rogue AI above present suffering; he calls Yudkowsky sincere and badly wrong. Two of his details are off: Singer’s The Life You Can Save is from 2009 (the 1970s piece is “Famine, Affluence, and Morality”), and Mira Murati was not on the board that fired Altman in 2023. He ends on OpenAI’s finances, and doubts it lasts in its current form past 2027.