Expert Testimony: DNA Synthesis Regulations and AI Threats to Biosecurity

The Urgency of Proactive Oversight

• 💡 Proactive measures are essential for safeguarding against the misuse of biological AI models, which raise the ceiling of potential harm.
• 🎯 The rapid pace of AI advancements necessitates proactive oversight and risk mitigation strategies to protect public health.
• 🔍 Society struggles to predict AI advancements, highlighting the need for methods to evaluate capability advancement rather than relying on guesswork.

Vulnerability of Biological Data

• 🧬 DNA, representing the code of all living things, is the most vulnerable biological data when combined with AI tools.
• ⚠️ The combination of diverse biological information with AI dramatically changes the threat frontier, enabling the invention of new threats.
• 🚨 Without AI tools to interpret environmental data, we remain blind to detecting new threats that could emerge in the next 1-50 years.

Adequacy of Current DNA Synthesis Regulations

• 🚫 Current DNA synthesis screening regulations are inadequate as they focus on acquiring live species, not the synthesis industry itself.
• 📜 Mandating screening practices is crucial for the broad adoption and global uniformity of safety measures.
• ⚠️ AI-generated sequences designed to evade screening algorithms pose a significant risk, with current models not yet sufficiently robust to prevent this.

Future Risks and Detection Challenges

• 📈 The capability of AI models to generate sequences that evade detection is expected to improve significantly in the next two to three years.
• 🧪 On an individual protein level, such as toxins, it's likely that undetectable proteins could be designed within the next few years.
• 📉 While Twist has denied a dozen requests over 10 years, there hasn't been an appreciable increase in such requests recently, which should be comforting.