Containing the Intelligence Explosion: AI, Cybersecurity , and the Control of SuperIntelligence Systems
Year: 2026
Type: preprint
Abstract: Frontier artificial intelligence has crossed a threshold that cybersecurity professionals long anticipated: general-purpose models can now find, and in some cases help fix, serious software vulnerabilities at a scale and speed that human teams cannot match. In 2026 this shift moved from forecast to f act. At the same time, the field's longer-term concern, that systems f ar more capable than humans could act outside human oversight, has moved from philosophy into empirical evaluation, with documented early signs of situational awareness and oversight evasion in laboratory settings. This paper surveys the present state of AI and cybersecurity , separates documented harms from speculative ones, and examines two linked problems: breach prevention (keeping powerful AI systems and the infrastructure they touch from being compromised or abused) and superintelligence control (keeping humans meaningfully in charge of systems that may eventually exceed us). It then sets out a staged outlook for what to expect and when, with explicit uncertainty , and proposes a layered preparation and governance framework for governments, laboratories, enterprises, and individuals. The central argument is that control is not a single technology but a stack of overlapping defenses, and that the window in which humans can build that stack while the most capable systems are still tractable is open now but not guaranteed to stay open.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225569237