Skip to content

Cyber Threat Exposure

Cyber Threat Exposure quantifies how AI capabilities amplify offensive cyber operations and shift the offense-defense balance toward attackers. Unlike theoretical risks, AI-assisted cyberattacks are already occurring at scale—2025 saw AI-powered attacks surge 72% year-over-year, with 87% of organizations reporting AI-driven incidents.

The September 2025 Anthropic disclosure marked a watershed: the first documented AI-orchestrated cyberattack where AI executed 80-90% of tactical operations autonomously, targeting approximately 30 global entities. AI-generated phishing now achieves 54% click-through rates versus 12% for non-AI phishing—a 4.5x effectiveness improvement.

MetricScoreNotes
Changeability35Difficult—cyber threats are pervasive with structural defense disadvantages
X-risk Impact55Moderate-high—primarily through infrastructure attacks enabling other risks
Trajectory Impact45Moderate—economic disruption and erosion of digital trust
Uncertainty50Well-documented threats but offense-defense balance remains contested

Risks:

Models:

Key Debates:

  • Could AI-enabled cyberattacks cause physical catastrophe through infrastructure attacks?
  • Will AI improve cyber defense more than offense, or create an offense-dominant world?
  • Does AI-enabled anonymity make cyber conflict more likely by enabling deniable attacks?

Ratings

MetricScoreInterpretation
Changeability35/100Somewhat influenceable
X-risk Impact55/100Meaningful extinction risk
Trajectory Impact45/100Significant effect on long-term welfare
Uncertainty50/100Moderate uncertainty in estimates