Fireside Chat, Interview, Other
Oege De Moor (XBOW) & Apoorv Agrawal (Altimeter): Hackers with GPUs Offensive Security in the AI Era
- The AI platform Expo is projected to outperform all human hackers on the HackerOne platform, remain at the top of the leaderboard, and achieve superhuman capabilities in penetration testing within one year.
- Within the next 12 months, a major AI-powered cyber incident is expected to occur, catalyzing a temporary period of chaos due to the world's inability to counter increased AI-driven threats.
- Adversarial nation states and criminal organizations are predicted to develop autonomous AI tools to forge unprecedented cyber attacks, eliminating skill shortages and enabling non-technical users to attack systems with unrestricted frequency.
- Automation will reduce the window between security patch publication and exploitation in unpatched systems from weeks to less than a few hours, forcing human security professionals to shift from technical testing to orchestrating targets and business logic rules.
- The industry anticipates that the chaos caused by AI threats will eventually transition into a stable environment where defenders regain the upper hand.
- While current code generation tools are viewed as overstated and insufficient for solving hard architectural problems, non-technical users will gain the ability to build functional applications via platforms like Replit, and personalized AI tutors will become globally accessible for one-on-one education.
- Human creativity and decision-making will remain essential for architectural design, conceptual work, and the most creative aspects of penetration testing, as AI swarm approaches are not expected to fundamentally outperform single agents on these tasks.
- Specific technologies include Suno, noted for its ability to generate music and lyrics from text, and Replit, which is identified as a strong ecosystem beyond simple code generation.
- AI hallucinations may evolve into a strategic feature for cybersecurity applications, enabling agents to test unexpected vectors such as JSON content types.