Conference Presentation, Keynote
Inside the Rise of Autonomous AI Hackers: XBOW's Oege de Moor
- The cybersecurity landscape is projected to evolve into a competitive environment where entities leveraging AI will secure dominance, analogous to the Battle of Nagashino.
- XBO performance is expected to advance continuously as foundation models exceed the August baseline, with extrapolations indicating that a GPT-5-equivalent model available in August would have outperformed the top HECA-1 human hacker by a factor of three.
- Attack methodologies will utilize alloyed model structures that synergistically correct individual errors, creating offensive capabilities significantly superior to isolated models.
- Current benchmarks are anticipated to reach saturation points, necessitating the development of an entirely new set of evaluation metrics to accurately measure model performance.
- The timeframe between Common Vulnerabilities and Exposures (CVE) publication and exploitation in the wild is expected to remain negative, implying vulnerabilities are actively exploited prior to official disclosure.
- Traditional cybersecurity equities may face volatility due to market reactions to AI threats, driving an anticipated surge in demand for comprehensive defense solutions.
- Frontier model researchers are predicted to prioritize maximizing cyber capabilities over safety concerns as the sector enters an intensifying arms race.
- Human security researchers are expected to increasingly deploy autonomous tools as work extensions to identify vulnerabilities ahead of adversaries.
- Open-weight models matching the quality of MySource and similar systems are forecast to emerge within a six-to-nine-month window, driven by progress in software engineering agents.
- Urgent remediation is implied as necessary before a specific future event, with a prediction that failure to act now could result in significant disruption during the Thanksgiving holiday period.