The Connected Threat Picture: A Graded Assessment of AI Risk from Cognitive Influence to Autonomous Action

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Assessments of AI risk often separate present harms from catastrophic futures and treat each danger in isolation. This paper sets out a connected threat picture in which persuasive systems, autonomous agents, adversarial actors, concentrated infrastructure and economic disruption interact with the capacity of people and institutions to respond. It brings cognitive security together with governance, corporate power, physical security, biological misuse, labour and welfare. Each specified claim is assessed against its evidence, exposure and horizon; severity, confidence, likelihood, reversibility and intervention capacity remain distinct. A developer's September 2026 incident report documents misuse in cyber operations, influence campaigns and weapons development, while its simulated drone-development case does not establish field deployment of a lethal autonomous weapon. The paper examines industry warnings about autonomy and superintelligence, the possible pathway from economic grievance to political violence, bounded biohybrid demonstrations and the critical skills on which cognitive defence depends. Documented occurrence is distinguished from experimental demonstration, projection and speculation. The argument is that risk changes at transitions: when influence acquires persistent memory and relational authority, when advice gains permission to act, and when recognition of harm outpaces an institution's capacity to intervene. The contribution is a cross-domain assessment that specifies who can see and interrupt developing harm in the time available. It additionally examines cumulative data fusion, information overload, autonomous swarms and quantum-enabled cryptographic risk as interacting rather than standalone pathways. AI also offers substantial benefits in science, medicine, accessibility and defence, so the argument is for benefit secured by safeguards, and the paper treats cognitive security, including the protection of individual cognitive agency, as a matter of national safety as well as individual welfare.