Key Takeaways
- Anthropic's own alignment science lead publicly estimated a greater than 10% chance AI causes human extinction within a decade—then confirmed his lab has no plan yet to prevent it. A 2024 survey of 2,778 AI researchers placed the median probability of catastrophic AI outcomes at approximately 5%, with individual expert estimates ranging from near zero to above 80%.
- The CrowdStrike 2026 Global Threat Report found AI-enabled adversary activity surged 89% year over year, with average cyberattack breakout time collapsing to 29 minutes—and a record fastest intrusion of just 27 seconds.
- A 2025 study confirmed that AI virology capabilities had already surpassed benchmarks experts predicted wouldn't be reached until after 2030, while a Microsoft-led red-teaming exercise demonstrated that generative AI tools can design dangerous proteins capable of evading biosecurity screening systems used by DNA manufacturers worldwide.
