DEF CON 33 - BiC Village - Petty Proteins When Molecules Go Rogue& Why Cyberbiosecurity - Tia Pope

The video explores the emerging field of cyber biosecurity, emphasizing the risks and opportunities in AI-driven protein engineering. The speaker highlights how AI tools like ESM and ProGPT2 can generate novel protein sequences rapidly, raising concerns about unintended harmful applications such as bioweapons. A key message is that non-traditional backgrounds can contribute meaningfully to this space, and that vigilance—through Google searches and public database validation—is essential. The speaker advocates for guardrails, working groups, and ethical AI use to mitigate risks before they escalate.

Key Takeaways:
• The convergence of AI and protein engineering enables rapid design of new proteins, but lacks safeguards against malicious use 11:00.
• AI models like ESM and ProGPT2 are open-source and accessible, allowing anyone to generate protein sequences, but lack guardrails to prevent harmful outputs 15:18.
• Proteins are critical biological agents whose functions can be dramatically altered by small mutations, posing risks in medicine, environment, and bioweapons 17:00.
• Real-world translation of AI-generated proteins has been validated in wet labs with 90% accuracy, showing potential for both good and bad applications 26:00.
• The speaker is developing models like "Craig" and "Debo" to detect risks and trace origins of AI-generated proteins, aiming to create early-warning tools 22:00.

Cyber biosecurity is a growing, high-stakes field where AI enables rapid innovation but also amplifies risks—requiring interdisciplinary collaboration, transparency, and proactive regulation to prevent misuse.

Sources:

• 11:00