AI Threats Affecting Society at Large

How AI Threats Appear

For society at large, AI-enabled threats manifest as systemic, diffuse harms that cannot be attributed to specific individuals or organizations:

• Epistemic degradation — Widespread AI-generated content that erodes the shared capacity to distinguish fact from fabrication, undermining the epistemic foundations of public discourse
• Trust erosion — Cumulative loss of public confidence in institutions, media, and interpersonal communications due to the prevalence of AI-generated synthetic content
• Power concentration — Structural accumulation of economic and informational power by entities controlling advanced AI systems, reducing competitive diversity and democratic accountability
• Social fragmentation — AI-driven recommendation and content generation systems that create incompatible information environments, fragmenting shared reality
• Existential and catastrophic risk — Speculative but evidence-informed concerns about advanced AI systems whose optimization targets diverge from collective human welfare

Society at large is the highest-threshold affected group value. It should only be used when harm genuinely diffuses across society and cannot be meaningfully captured by other group values.

---

Relevant AI Threat Domains

• Systemic Risk — Infrastructure dependency, strategic misalignment, and uncontrolled capability escalation
• Information Integrity — Large-scale erosion of public trust in information ecosystems
• Economic & Labor — Structural market concentration and economic power asymmetry
• Human-AI Control — Gradual transfer of societal decision-making to AI systems

---

What to Watch For

Indicators of AI-related societal-level harm:

• Measurable decline in public trust in information sources, institutions, or democratic processes correlated with AI content proliferation
• Market concentration metrics showing AI-driven consolidation across multiple sectors
• Evidence of AI systems influencing public opinion or behavior at population scale
• Increasing difficulty in attributing authorship or verifying authenticity of public communications
• Cascading effects where AI failures in one domain propagate to create systemic instability

---

Regulatory Context

• EU AI Act — Establishes systemic risk assessment requirements for general-purpose AI models
• NIST AI RMF — Addresses societal-scale AI risks through organizational risk management
• International AI governance initiatives (UN, OECD, G7) address cross-border systemic AI risks
• AI safety research institutions (MIRI, ARC, CAIS) contribute to understanding existential and catastrophic AI risk trajectories

For classification rules and evidence standards, refer to the Methodology.