NeurIPS 2025 Workshop

Biosecurity Safeguards for Generative AI

San Diego, California, USA

Schedule (Tentative)

Single-track • All times PDT. Invited talks = 25 min + 3–5 min Q&A. Contributed spotlights = 5 min (strict). Panel = 45 min. Posters = 40 min.

Time (PDT) Session Speaker(s)
9:00 – 9:05Opening remarksOrganizers
9:05 – 9:35Invited talk 1Yoshua Bengio (Mila, Université de Montréal)
9:35 – 10:05Invited talk 2Peter Henderson (Princeton University)
10:05 – 10:35Invited talk 3Le Cong (Stanford)
10:35 – 10:45Short break—
10:45 – 11:15Invited talk 4Megan Blewett (Iris Medicine; Kelonia Therapeutics)
11:15 – 11:45Invited talk 5Sandy Yao
11:45 – 12:00
Contributed Spotlights A (3 × 5 min)
  • 11:45–11:50 ABC-Bench: An Agentic Bio-Capabilities Benchmark for Biosecurity
  • 11:50–11:55 Without Safeguards, AI-Biology Integration Risks Accelerating Future Pandemics
  • 11:55–12:00 Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
 —
12:00 – 1:00Lunch break—
1:00 – 1:40Poster session 1—
1:40 – 2:10Invited talk 6Jian Ma (Carnegie Mellon University)
2:10 – 2:40Invited talk 7Russ Altman (Stanford University)
2:40 – 3:25 Panel discussion (45 min) Panelists
3:25 – 3:35Short break—
3:35 – 4:05Invited talk 8Eric Xing (Carnegie Mellon University; MBZUAI; Petuum Inc.)
4:15 – 4:30
Contributed Spotlights B (3 × 5 min)
  • 4:15–4:20 GeneBreaker: Jailbreak Attacks against DNA Language Models with Pathogenicity Guidance
  • 4:20–4:25 Resisting RL Elicitation of Biosecurity Capabilities: Reasoning Models Exploration Hacking on WMDP
  • 4:25–4:30 SafeGenie: Erasing Dangerous Concepts from Biological Diffusion Models
 —
4:30 – 5:30Poster session 2—
5:30 – 5:35Closing remarksOrganizers

The workshop runs roughly 9:00–5:35 pm PDT with invited talks, two contributed spotlight blocks, a panel, and two poster sessions.