He works at OpenAI: his job is to stop ChatGPT from helping make a Bioweapon

Summary
Juan Felipe Cerón Uribe, an AI adversarial robustness researcher at OpenAI who works on mitigating bioweapon risks in ChatGPT, is interviewed by Eli Tyre of Palisade Research. Cerón Uribe reflects on internal organizational pressures to race ahead, the plausibility of severe or existential risks, and why frontier labs and governments must coordinate to pace advanced AI development.

What is shown

• [00:00] Teaser remarks on the catastrophic risks of unmonitored AI acceleration and engineered biological threats.
• [00:33] Juan Felipe Cerón Uribe introduces himself, noting his start at OpenAI in 2022 working on what became ChatGPT, his current focus on bioweapon risk mitigation, and clarifying that he speaks in his personal capacity.
• [01:05] Reading from his phone a prepared statement on the extreme upsides versus catastrophic risks of frontier AI, and how labs race forward blindly.
• [02:56] Discussion of corporate and managerial pressures, employee incentives, and internal celebrations of larger models driving acceleration.
• [07:00] Explanation of the multi-layered defense stack used to prevent models from assisting with biological weapons (model refusal training, separate monitor models reading outputs, and input/output classifiers).
• [09:30] Discussion of the concept of “pacing the frontier,” arguing that pacing must be internationally coordinated (notably between the US and China) rather than unilateral.
• [14:52] Reflection on internal culture shifts at OpenAI between 2022 and 2026, noting that conversations about existential risk have decreased as the company became larger, more commercial, and product-focused.
• [19:05] Discussion of recursive self-improvement (AI models conducting automated AI research) and potential bottlenecks like GPU compute.
• [22:32] Proposals for societal safeguards beyond AI pacing, including screening biological reagent orders, verifying DNA synthesis customers, and hardening cybersecurity defenses.
• [28:47] Recommendations for policymakers to prioritize safety over speed and avoid “tossing a coin” on human survival.
• [33:00] Outro by interviewer Eli Tyre outlining Palisade Research’s mission to present perspectives from frontier AI laboratory employees.

Claims & numbers

• Cerón Uribe states he joined OpenAI as a research engineer in 2022 and worked on the development of ChatGPT (presenter says).
• He claims OpenAI, Anthropic, and other frontier developers rely on a multi-component stack including refusal post-training and secondary monitoring models to detect and block dual-use biological weapon queries (presenter says).
• He notes that discussions about existential risk at OpenAI were common in 2022 (e.g., at lunch tables) but have steadily decreased as company focus shifted toward shipping products, revenue, and commercial competition with Anthropic’s Claude (presenter says).
• He states that unilateral pauses would fail because competitor labs and other nations (such as China) would continue advancing, necessitating international government-level coordination (presenter says).

Notable quotes

• [01:24] “Instead, frontier developers are racing blindfolded and crossing their fingers.”
• [09:36] “Pacing the frontier for me means you slow the development of advanced AI until you’re confident that when you take that next step in development, you’re not going to cause a catastrophe.”
• [29:04] “If heads is cure cancer and tails is we all die, we shouldn’t be tossing that coin.”

Assessment
This video is a genuine, long-form sit-down interview conducted by Palisade Research as part of their series exploring insider perspectives on frontier AI safety. The speaker shares personal observations, technical philosophies, and workplace culture insights without displaying proprietary source code or confidential company materials.

Described by gemini-3.8-flash on 2026-10-15 from the video’s audio and frames.