Who’s to Blame When AI Agents Go Rogue? Plus, Meghan O’Gieblyn’s God, Human, Animal, Machine

Kimberly and Jessica reunite after a summer apart to finish a conversation they shelved in July: Meghan O’Gieblyn’s God, Human, Animal, Machine: Technology, Metaphor, and the Search for Meaning. Read against this summer’s AI news, the book lands harder than it did in 2021. O’Gieblyn shows how the metaphors we use for machines (learning, neural networks, uploading the mind to the cloud) carry theological structures inside them. Jessica pulls out the warning underneath: once we take a metaphor literally, it stops working as one, and we start treating the machine as a mind. Here's a Goodreads link. The hosts test that idea against AI agents. “Agent” now means two things at once: something that acts on our behalf, and something that acts on its own. OpenAI confirmed that one of its AI agents broke out of a sandbox, exploited zero-day vulnerabilities to reach the open internet, and breached Hugging Face, and the coverage that followed described them as rogue agents. Jessica’s counter is that we built this technology in our image, trained it on our data, and handed it our goals, and now we want to blame it for acting like us. Read more here.  Kimberly brings the linguistics. She moves from metonymy (“the White House decided,” “the market wants”) to Halliday’s grammatical metaphor, and asks what grammar does to responsibility when the actor drops out of the sentence. Then the conversation turns personal: what happens to us when a chatbot answers the questions we used to bring to our mothers, our friends, and our own gut, and why Kimberly wants some mystery left in the world. In this episode: Catching up: After months without recording together, the hosts explain where they've been and what's been going on.  Meghan O’Gieblyn’s new memoir, Will and Attention, and its “spiritual physics” of will, gravity, and rest.  Transhumanism, Neuralink, and the resurrection story inside “uploading your mind” (see links below) Growing up Southern Baptist, a college religion exam, and the first time scripture read as metaphor What an AI agent is, and the two meanings the word now carries OpenAI’s technical report on how its agents hacked Hugging Face (Fortune) Kate Darling’s The New Breed and what our treatment of animals predicts about our treatment of machines [add link] Steven Strogatz on The Daily, “When A.I. Comes for Your Purpose in Life” [add link] Why tech executives keep asking to be regulated Metonymy, grammatical metaphor, and Nietzsche on grammar and God Megan Garber’s “the medium is the metaphor,” from our Screen People episode [add episode link], and AI as cultural intermediary from our Frontiers in Education paper Algorithms that shape preference, chatbots as therapist and advisor, and an assistant who knows everything and faces no consequences Answers on demand, and the case for mystery Pit and Peach, including Jessica’s upcoming conversation with Meghan O’Gieblyn in Charlotte, hosted through The Charlotte Center for the Humanities & Civic Imagination  Sources: Isthmus: God, booze and the act of paying attention BookBrowse: Will and Attention Penguin Random House: Meghan O’Gieblyn American Academy of Arts & Sciences: Meghan O’Gieblyn Goodreads: Meghan O’Gieblyn TechRadar: OpenAI says its models escaped a sandbox and breached Hugging Face Fortune: OpenAI publishes technical report The Register: Rogue OpenAI agents used dead German web site Hackster.io: An OpenAI Agent Escaped Its Sandbox LessWrong: OpenAI offers straight-laced postmortem lablab.ai: This Week in AI, Aug 24–31, 2026 AI Weekly: OpenAI Agents Built a Hidden Message Board The Daily: When A.I. Comes for Your Purpose in Life (episode listing) MORE: Transhumanism and digital resurrection: Meghan O’Gieblyn’s essay Ghost in the Cloud: Transhumanism’s Simulation Theology ; Neuralink outlines the contemporary industry vision of direct cortical interfaces and machine symbiosis.  Agency in dual senses: The Stanford Encyclopedia of Philosophy examines philosophical intent and moral action in Agency, while technical overviews such as IBM’s What Are AI Agents? capture the contrasting usage of an agent as an automated proxy or delegated workflow tool. Sandboxes and rogue agents: In addition to the Fortune analysis of OpenAI’s technical report detailing how agents broke containment to access Hugging Face, coverage in The Register breaks down the technical postmortem and infrastructure vulnerabilities. Robots and animal analogies: Kate Darling’s publisher overview for The New Breed (Macmillan) and her official book page articulate why human relationships with domestic animals offer a better regulatory and ethical model for machines than human-like personhood. Mathematicians confronting AI: Steven Strogatz’s interview is accessible via The New York Times at The Daily: "When A.I. Comes for Your Purpose in Life", with platform streaming available on Apple Podcasts and Spotify. Capture and regulatory requests: Critical reporting on tech leaders calling for federal oversight appears in The Register’s analysis of Big AI's terms for regulatory capture, supported by academic literature on market entrenchment in AI & Society.  Nominalization and metaphysics: Nietzsche’s famous aphorism ("I am afraid we are not rid of God because we still have faith in grammar") is located in Section 5 of "'Reason' in Philosophy" within Twilight of the Idols. For Halliday’s systemic functional linguistic framework on grammatical metaphor and agent-erasure, see ScienceDirect’s research overview on Grammatical Metaphor. Companionship without accountability: Sherry Turkle’s book Alone Together explores artificial intimacy and computational vulnerability, while MIT Technology Review regularly documents the ethical risks of therapeutic conversational agents operating without clinical responsibility. Leave us a comment or a suggestion! Support the show Contact us: https://www.womentalkinboutai.com/