The Skeptics' Guide to the Universe The Skeptics Guide #1084 - Apr 18 2026
9 snips
Apr 18, 2026 A lively rundown of Artemis II's translunar tests, heat‑shield performance and improvised crew shielding. Deep dives into AI-driven biology labs, dual‑use risks and the policy gaps around bio‑AI. New automated coding tools that can find software vulnerabilities and coordinated white‑hat responses. A proof‑of‑concept superconductivity imaging breakthrough and skeptical takes on red light therapy hype.
AI Snips
Chapters
Transcript
Episode notes
AI Greatly Expands Dual Use Risks In Biology
- AI lowers the expertise barrier for biological design, enabling scientifically naive users to attempt producing dangerous pathogens or optimize viral spread.
- Studies show LLMs can enhance protein evolution, optimize viral transmissibility, and guide recovery of live viruses from synthetic DNA.
Update Governance Not Just Industry Promises
- Avoid relying solely on industry self-regulation for AI biosecurity; pursue targeted surgical interventions in policy and oversight.
- Update treaties and governance (e.g., Biological Weapons Convention) to explicitly address AI-enabled biological design and automation.
Powerful Coding AI Can Become An Automated Hacker
- Anthropic's Mythos coding model can both generate code and find/exploit deep vulnerabilities across widely used software, creating an unprecedented cybersecurity threat.
- The company created Project Glasswing, sharing a restricted Mythos Review with 40 firms to proactively find and fix vulnerabilities.