EA Forum Podcast (All audio) “When is it important that open-weight models aren’t released? My thoughts on the benefits and dangers of open-weight models in response to developments in CBRN capabilities.” by Ryan Greenblatt
Jun 11, 2025
Ryan Greenblatt, an insightful author from the Effective Altruism Forum, dives into the contentious topic of open-weight AI models and their potential to assist in developing CBRN weapons. He discusses recent concerns from Anthropic regarding models that could empower individuals with basic STEM knowledge to create bioweapons. This raises crucial questions about the ethical implications and necessary governance of such technologies. Greenblatt also reflects on his views regarding the release of models with these capabilities and possible mitigations.
AI Snips
Chapters
Transcript
Episode notes
Open-Weight Models' Cost-Benefit Tradeoff
- Open-weight AI models capable of aiding amateurs with bioweapons could cause about 100,000 expected fatalities yearly.
- These models, however, also reduce larger existential risks, making their release a complex trade-off.
Balancing AI Takeover Risks
- Large AI risks, like loss of control, overshadow direct harm caused by CBRN capabilities in open models.
- These risks likely make the benefits of open-weight models outweigh their direct fatality costs.
Advocate Safety and Transparency
- Pressure AI companies to uphold safety commitments and not mislead about system dangers.
- Advocate for transparency when companies weaken safety policies related to dangerous capabilities.