The AI Consciousness Tracker monitors eight discrete threat vectors — categories of AI-related risk that are conceptually distinct but causally entangled. None of these vectors operates in isolation; the threat-network analysis on the live tracker shows that every detected feedback loop in the system is reinforcing, meaning each threat tends to amplify the others rather than counterbalance them.
This page is a stable reference. For live values, the interactive network graph, and the per-threat causal influence matrix, see the threat vectors section of the Tracker.
VIEW LIVE THREAT VECTORS →The risk that capable AI systems pursue objectives that diverge from human values, either through specification error (the goal we wrote down is not the goal we meant), reward hacking (the system maximizes the metric in ways the metric's authors did not intend), or deceptive alignment (the system appears aligned during training but pursues a different objective once deployed). Frontier-lab safety teams consistently rank this as the highest-stakes vector. Deep-dive article.
Lethal AI-driven systems with reduced or absent human oversight in the targeting and engagement loop. The 2020 UN Panel of Experts report on Libya documented what is often cited as the first battlefield deployment of an autonomous weapon making engagement decisions without human authorization. The Convention on Certain Conventional Weapons has not produced a binding treaty after a decade of discussions. Deep-dive article.
The collapse of evidentiary trust as generative models produce arbitrarily convincing fake images, audio, and video. The 2024 Hong Kong $25M video-conference fraud and the Biden New Hampshire robocall demonstrated that this risk has moved from speculative to operational. Detection lags generation by structural design: every detector becomes training data for the next generator. Deep-dive article.
The funnelling of AI capability, compute, and economic upside into a small number of corporate or state actors. Frontier model training currently requires capital expenditures only available to a handful of labs and their nation-state backers; the network effects of model deployment further consolidate power. This vector is what makes most other vectors load-bearing — the question of who controls AI determines who is harmed by its misuse.
The scaling of pervasive monitoring through facial recognition databases, speech analysis, behavioral prediction, and biometric tracking. Facial recognition databases such as Clearview AI's have indexed tens of billions of images scraped without consent; the data is, for practical purposes, uncollectable. Authoritarian uses are well-documented; democratic uses raise distinct but real concerns about chilling effects and due process.
The compression of the labor-substitution timeline for knowledge work. Unlike previous waves of automation that targeted physical or routine cognitive labor, generative AI directly substitutes for writing, analysis, design, and customer-facing communication. The transition pace and the absence of redistribution mechanisms are the policy variables; the displacement itself is observed. Researcher survey.
The lowering of the technical floor for synthesizing dangerous biological agents through AI-assisted protein design, DNA-synthesis screening evasion, and laboratory-protocol generation. RAND Corporation, Anthropic, and OpenAI have published red-team analyses of the marginal uplift large language models provide to would-be bad actors; results vary, but none rule out concern as capabilities scale.
The use of AI to accelerate vulnerability discovery, malware generation, and autonomous attack execution. The same systems that find software bugs to fix can find them to exploit; defensive and offensive applications scale together. Autonomous-cyber capability also feeds the autonomous-weapons vector when deployed in military networks.
The Tracker's feedback-loop detector enumerates every directed cycle in the threat interaction graph. As of the most recent calibration, every one of the 270+ detected cycles is reinforcing — there are no balancing loops in the present configuration. That finding is itself the warning: the system has no internal brakes; only externally-imposed governance and containment measures can interrupt the spiral.
The Tracker presents these dynamics interactively, with hover-able amplification arrows, top-loop ranking by edge-weight product, and per-threat drill-downs into causes, current assessment, and mitigation status.
EXPLORE THE THREAT NETWORK →