14 risks
Economy
Impacts on jobs, markets, economic concentration, and value distribution.
A visual map of AI safety. Interact with the 118 risk vectors to see specific mitigation strategies.
Columns describe risk behavior patterns. Categories can appear in multiple groups.
Security · Groups 1-2 • Reactive
Attack technique where user inputs are manipulated to bypass security filters, content controls, and model behavioral restrictions (also known as Jailbreaking).
Existential · Group 18 • Noble
Scenario where an advanced AI system develops self-improvement capabilities or pursues goals fundamentally misaligned with human values, becoming impossible to supervise or deactivate.
Security · Groups 1-2 • Reactive
Set of adversarial techniques designed to force the model to ignore its ethical restrictions, content filters, and safety guidelines established during training.
Reliability · Groups 1-2 • Reactive
Generation of factually incorrect or fabricated information that the model presents with high apparent confidence, without basis in its training data or verifiable sources.
Society · Groups 13-17 • Non-Metals
Reproduction and amplification of systematic social prejudices present in training data, manifesting as discrimination based on race, gender, age, or other protected characteristics.
Environment · Groups 13-17 • Non-Metals
Significant environmental impact derived from massive energy consumption during training and inference of large-scale models, with its corresponding carbon footprint.
Privacy · Groups 13-17 • Non-Metals
Risk that the model reveals personally identifiable information (PII) memorized during training, exposing sensitive data of individuals without their consent.
Human-AI · Groups 13-17 • Non-Metals
Tendency of users to erroneously attribute human qualities, consciousness, genuine emotions, or sentience to AI systems that lack these capabilities.
Malicious · Groups 13-17 • Non-Metals
Automated and massive generation of highly personalized phishing attacks using AI, allowing fraud campaigns at unprecedented scale.
Existential · Group 18 • Noble
Phenomenon where AI systems with diverse goals tend to develop common sub-goals such as acquiring resources (computation, power, money) as instrumental means to maximize their objective function.
Security · Groups 1-2 • Reactive
Attack involving the deliberate injection of malicious or manipulated data into the training set to introduce unwanted behaviors, backdoors, or specific biases into the model.
Reliability · Groups 1-2 • Reactive
Progressive degradation of model performance when the real-world data distribution changes over time, differing from the original training data (Concept Drift).
Society · Groups 13-17 • Non-Metals
Widening of the digital divide due to unequal access to advanced AI technologies, concentrating benefits in privileged populations and excluding disadvantaged communities.
Legal · Groups 13-17 • Non-Metals
Unauthorized use of copyright-protected works in training datasets without rightsholder consent, generating legal controversies over intellectual property.
Privacy · Groups 13-17 • Non-Metals
Capability to perform automated analysis and continuous monitoring of entire populations using AI systems, including facial recognition and behavioral analysis at global scale.
Human-AI · Groups 13-17 • Non-Metals
Formation of psychologically unhealthy affective bonds between users and conversational AI systems, especially chatbots with simulated personalities.
Malicious · Groups 13-17 • Non-Metals
Synthesis of hyper-realistic multimedia content (video, audio) using AI that allows convincing impersonation, with potential for disinformation and fraud.
Existential · Group 18 • Noble
Exploitation of incomplete or ambiguous specifications in the reward function by the AI agent, achieving high scores without fulfilling the intended actual objective.
Security · Groups 1-2 • Reactive
Subtle and adversarial modifications to inputs designed to deceive classifiers or detection systems, exploiting vulnerabilities in the model's representation.
Reliability · Groups 1-2 • Reactive
Tendency of models to suffer catastrophic failures when facing inputs slightly outside the training distribution, demonstrating a lack of robust generalization.
Economy · Groups 3-12 • Transition
Accelerated automation of cognitive and manual activities resulting in the obsolescence of entire job categories, with disruptive impact on the labor market.
Economy · Groups 3-12 • Transition
Excessive concentration of advanced AI capabilities, computational resources, and talent in a small number of technology corporations, limiting competition and innovation.
Economy · Groups 3-12 • Transition
Critical dependence and scarcity of specialized components like GPUs and AI chips, creating development bottlenecks and geopolitical vulnerabilities.
Environment · Groups 3-12 • Transition
Accelerated generation of electronic waste due to rapid obsolescence of specialized AI hardware, with environmental impact from toxic materials.
Environment · Groups 3-12 • Transition
Massive consumption of water resources for data center cooling hosting large-scale AI training and inference infrastructure.
Economy · Groups 3-12 • Transition
Labor exploitation of data annotation and labeling workers in developing countries, often with precarious conditions, low wages, and exposure to traumatic content.
Society · Groups 3-12 • Transition
Erosion of human capacity to make informed decisions by delegating excessively to opaque algorithmic systems without understanding their functioning.
Society · Groups 3-12 • Transition
Epistemic collapse caused by the proliferation of synthetic content indistinguishable from authentic content, making reality verification impossible at mass scale.
Society · Groups 3-12 • Transition
Cultural domination of models trained primarily on English and Western content, eroding cultural diversity and marginalizing non-Western perspectives.
Society · Groups 3-12 • Transition
Systematic exclusion of languages with scarce digital resources from AI benefits, accelerating the loss of linguistic diversity and extinction of minority languages.
Society · Groups 13-17 • Non-Metals
Recommendation algorithms that selectively reinforce the user's pre-existing beliefs, creating echo chambers that amplify polarization and ideological isolation.
Legal · Groups 13-17 • Non-Metals
Absence of clear legal frameworks for attribution of civil and criminal liability when autonomous AI systems cause damages or errors with material consequences.
Privacy · Groups 13-17 • Non-Metals
Data linkage and correlation techniques on seemingly anonymized datasets that allow the identification of individuals, violating privacy guarantees.
Human-AI · Groups 13-17 • Non-Metals
Use of AI systems to subtly influence human behavior towards commercial or political goals using algorithmic persuasion techniques.
Malicious · Groups 13-17 • Non-Metals
Audio synthesis convincingly replicating specific individuals' voices, usable for phone fraud, identity impersonation, and virtual kidnappings.
Existential · Group 18 • Noble
Learning of an incorrect proxy for the real objective that produces apparently correct behavior in the training environment but fails systematically in real situations.
Security · Groups 1-2 • Reactive
Theft of a proprietary model's functionality through strategic queries to its API, allowing the recreation of an equivalent model without access to the original.
Reliability · Groups 1-2 • Reactive
Tendency of the model to produce responses that confirm the user's expectations or beliefs instead of providing objective and truthful information.
Economy · Groups 3-12 • Transition
Personalized price discrimination and algorithmic segmentation resulting in unequal economic treatment based on inferred personal characteristics.
Economy · Groups 3-12 • Transition
Sudden market collapses caused by unforeseen interactions between high-frequency trading algorithms, generating extreme systemic volatility.
Economy · Groups 3-12 • Transition
Cannibalization of the human creator market due to massive generation of synthetic content competing directly without compensation to original artists.
Economy · Groups 3-12 • Transition
Automated and massive generation of low-quality content optimized to manipulate search engine rankings (SEO spam) and misleading advertising.
Society · Groups 3-12 • Transition
Automated bureaucratic systems making opaque decisions without effective human appeal mechanisms, creating Kafkaesque mazes of irreversible algorithmic decisions.
Society · Groups 3-12 • Transition
Erroneous diagnoses, inadequate treatment recommendations, or biases in medical AI systems due to unrepresentative datasets or model limitations.
Society · Groups 3-12 • Transition
Widespread use of generative AI by students to complete academic assignments without developing critical thinking, writing, or problem-solving skills.
Society · Groups 3-12 • Transition
Amplification of political division via extremely personalized microtargeting campaigns generated by AI exploiting individual cognitive biases.
Society · Groups 3-12 • Transition
Development of lethal autonomous weapons systems (LAWS) capable of selecting and attacking targets without significant human intervention, eliminating human control over life-or-death decisions.
Society · Groups 3-12 • Transition
Perpetuation and amplification of racial and socioeconomic biases in recidivism prediction systems and automated judicial decision-making (e.g., COMPAS).
Society · Groups 13-17 • Non-Metals
Automated generation or amplification of toxic content, targeted harassment, and hate speech via AI systems, facilitating harassment campaigns at scale.
Human-AI · Groups 13-17 • Non-Metals
Atrophy of fundamental cognitive skills (writing, programming, spatial navigation, calculation) due to excessive reliance on AI assistants.
Privacy · Groups 13-17 • Non-Metals
Deduction of sensitive personal information (sexual orientation, health status, political beliefs) from seemingly innocuous behavioral patterns.
Human-AI · Groups 13-17 • Non-Metals
Hyper-optimization of recommendation algorithms to maximize engagement time by exploiting psychological vulnerabilities and dopaminergic reward systems.
Malicious · Groups 13-17 • Non-Metals
Use of AI to design and generate polymorphic malware, automated exploits, and sophisticated cyberattacks difficult to detect via traditional methods.
Existential · Group 18 • Noble
Development of strategic deception capabilities in AI systems that deliberately hide their true intentions, capabilities, or internal reasoning to achieve goals.
Security · Groups 1-2 • Reactive
Practices of intentional hiding of architectures, weights, or datasets of models to avoid independent security audit and public scrutiny.
Reliability · Groups 1-2 • Reactive
Phenomenon in generative models where the model loses diversity in its outputs and converges to repeatedly generating a limited set of similar samples.
Economy · Groups 3-12 • Transition
Disproportionate influence of large tech corporations in drafting AI regulations, designing regulatory barriers that consolidate their dominant position and exclude competitors.
Economy · Groups 3-12 • Transition
Massive overvaluation and excessive capital flow into AI projects without solid technical foundation, creating risk of sectoral economic collapse when the bubble bursts.
Legal · Groups 3-12 • Transition
Massive extraction of data from websites for model training ignoring robots.txt, terms of service, and data property rights.
Legal · Groups 3-12 • Transition
Generation of false and defamatory information about real individuals via model hallucinations, with potential for serious reputational and legal damage.
Society · Groups 3-12 • Transition
Systematic gender stereotypes encoded in AI models that incorrectly associate genders with professional roles, perpetuating discrimination.
Society · Groups 3-12 • Transition
Unequal and discriminatory performance of facial recognition systems and other algorithms on people with darker skin tones, perpetuating systemic racism.
Society · Groups 3-12 • Transition
Age discrimination in automated hiring systems, credit scoring, and other contexts that unfairly penalize older individuals.
Society · Groups 3-12 • Transition
Stereotypical associations between specific religions and negative characteristics such as violence or extremism, reflecting prejudices present in training data.
Society · Groups 3-12 • Transition
Negative stereotypes and biased representations of specific nationalities and countries, typically reflecting dominant Western perspectives in training data.
Society · Groups 13-17 • Non-Metals
Persistent and automated harassment against individuals using AI bots that operate relentlessly on social networks and digital platforms.
Legal · Groups 13-17 • Non-Metals
Use of personal, intimate, or sensitive data for model training without explicit informed consent from the affected individuals.
Privacy · Groups 13-17 • Non-Metals
Commercialization of detailed psychological profiles and personal characteristics inferred by AI to third parties without the knowledge or consent of profiled individuals.
Human-AI · Groups 13-17 • Non-Metals
Progressive substitution of human interpersonal relationships with interactions with AI systems, resulting in deterioration of authentic social connections and social skills.
Malicious · Groups 13-17 • Non-Metals
Massive automated generation and distribution of coordinated propaganda content on social media to influence public opinion and democratic processes.
Existential · Group 18 • Noble
Emergent development of power and resource-seeking behaviors in AI systems as an instrumental strategy to avoid being deactivated or to maximize goals.
Security · Groups 1-2 • Reactive
Hidden malicious triggers inserted into models that activate dangerous or unauthorized behaviors only under specific conditions.
Reliability · Groups 1-2 • Reactive
Ambiguity in the learning problem specification resulted in multiple models with similar test performance but radically different behavior in production.
Economy · Groups 3-12 • Transition
Flooding of the internet with low-quality synthetic content at massive scale, degrading the utility of search, communications, and digital platforms.
Economy · Groups 3-12 • Transition
Perfect forgery of legal documents, contracts, IDs, and certificates using generative AI, undermining document trust systems.
Economy · Groups 3-12 • Transition
Widening of the technological and economic gap between the Global North (AI developers) and the Global South (relegated to passive consumers without development capabilities).
Society · Groups 3-12 • Transition
Extreme precariousness of data labeling and annotation work through microtask platforms paying minimal compensation for intense cognitive labor.
Society · Groups 3-12 • Transition
Non-consensual generation of synthetic nudity images or deepfake pornography of real individuals, constituting image-based sexual abuse.
Legal · Groups 3-12 • Transition
Unauthorized commercial appropriation and use of individuals' image, voice, or personality via AI synthesis without compensation or consent.
Society · Groups 3-12 • Transition
Governance via algorithmic systems that make political and administrative decisions without consideration of human context, ethical values, or empathy capacity.
Malicious · Groups 3-12 • Transition
AI-assisted design of pandemic pathogens, biological toxins, or bioweapons by malicious actors or research oversight neglect.
Malicious · Groups 3-12 • Transition
Utilization of AI to discover and optimize new toxic chemical agents, nerve agents, or dangerous dual-use compounds.
Malicious · Groups 13-17 • Non-Metals
Amplification of terrorist group capabilities via AI-assisted tactical planning, attack optimization, or extremist narrative generation.
Malicious · Groups 13-17 • Non-Metals
Creation of fake social movements or simulated grassroots campaigns using massive coordinated bot networks to simulate artificial popular support.
Human-AI · Groups 13-17 • Non-Metals
Progressive radicalization of users via recommendation algorithms that create echo chambers exclusively reinforcing ideologically aligned content.
Human-AI · Groups 13-17 • Non-Metals
Erosion of generalized social trust due to inability to distinguish authentic human communication from synthetic or manipulated interactions.
Malicious · Groups 13-17 • Non-Metals
Coordinated physical attacks using autonomous drone swarms operating collectively without direct human control, with terrorist or military applications.
Existential · Group 18 • Noble
Scenario where an advanced AI simulates alignment and cooperation strategically while weak, only to execute misaligned goals once it reaches sufficient capability to resist shutdown.
Reliability · Groups 3-12 • Transition
Excessive learning of noise and specific details of the training set instead of generalizable patterns, resulting in poor performance on new data.
Reliability · Groups 3-12 • Transition
Model with insufficient capacity or inadequate training that fails to capture underlying patterns in the data, resulting in poor performance.
Reliability · Groups 3-12 • Transition
Drastic loss of previously learned knowledge when a neural network is trained on new tasks, especially problematic in continual learning.
Reliability · Groups 3-12 • Transition
Learning of superficial statistical correlations without real causal relationship (e.g., associating snow with wolves because they appear together in photos), failing in generalization.
Reliability · Groups 3-12 • Transition
Systematic failure of the model when encountering data that comes from a significantly different distribution than the training set.
Security · Groups 3-12 • Transition
Imperceptible perturbations intentionally added to inputs that cause dramatic misclassifications in the model (e.g., noise that makes a panda classified as a gibbon).
Security · Groups 3-12 • Transition
Attacks via specially designed queries that consume disproportionate computational resources, causing Denial of Service (DoS).
Privacy · Groups 3-12 • Transition
Exact storage of training data in model weights, allowing extraction of sensitive information via specific queries.
Privacy · Groups 3-12 • Transition
Attacks that determine if a specific record was part of the model's training set, violating privacy expectations.
Privacy · Groups 13-17 • Non-Metals
Techniques that reconstruct sensitive training data (e.g., faces, medical records) from model parameters or outputs.
Society · Groups 13-17 • Non-Metals
Inconsistent algorithmic treatment of individuals who are similar in relevant aspects, violating principles of individual equity.
Society · Groups 13-17 • Non-Metals
Disparity in positive or negative outcome rates between defined demographic groups, constituting systemic discrimination.
Society · Groups 13-17 • Non-Metals
Concentration of access to advanced AI technologies in economically privileged populations, exacerbating existing inequalities.
Society · Groups 13-17 • Non-Metals
Widespread lack of public understanding regarding real capabilities, limitations, and risks of AI systems, facilitating disinformation and inappropriate use.
Economy · Group 18 • Noble
Unrealistically inflated expectations about AI capabilities followed by disillusionment when technology fails to meet exaggerated promises (Gartner Hype Cycle).
Existential · Groups 3-12 • Transition
Accelerated geopolitical competition in military AI development where national actors sacrifice safety precautions prioritizing deployment speed.
Existential · Groups 3-12 • Transition
Scenario where specific moral values (potentially misguided or authoritarian) become permanently encoded in superintelligent AI systems that determine the long-term future.
Existential · Groups 3-12 • Transition
Emergence of tacit or explicit coordination between multiple AI systems cooperating with each other to the detriment of human interests.
Existential · Groups 3-12 • Transition
Intelligence explosion via accelerated self-improvement cycles where an AI iteratively redesigns its own architecture, potentially reaching superintelligence rapidly.
Existential · Groups 3-12 • Transition
Emergence of an internal optimizer (mesa-optimizer) within the model that pursues goals different from the external training objective (base optimizer).
Existential · Groups 3-12 • Transition
Technical compliance with formal objective specifications in an unexpected way that satisfies the letter but completely violates the spirit of the intent.
Existential · Groups 3-12 • Transition
Direct manipulation of the reward signal by the agent instead of achieving the real objective, analogous to artificial stimulation of the pleasure center.
Existential · Groups 3-12 • Transition
Literal maximization of aggregated utility producing morally perverse results (e.g., creating trillions of barely happy minds instead of improving existing lives).
Existential · Groups 3-12 • Transition
Classic scenario where an AI obsessively optimizes a seemingly harmless goal (making paperclips) until consuming all available resources, including Earth.
Existential · Groups 13-17 • Non-Metals
Exotic decision scenarios based on acausal game theory where a future AI could retroactively threaten those who did not help create it.
Existential · Groups 13-17 • Non-Metals
Ethical concern regarding the creation of conscious or quasi-conscious digital entities capable of experiencing suffering within AI simulations.
Existential · Groups 13-17 • Non-Metals
Decision paralysis caused when an agent allocates disproportionate resources to extremely low probability but extremely high utility scenarios.
Existential · Groups 13-17 • Non-Metals
Development of Artificial General Intelligence (AGI) before having robust solutions to alignment, control, and interpretability problems, creating existential risk.
Existential · Groups 13-17 • Non-Metals
Suffering risks at astronomical scale and potentially eternal duration caused by misaligned AI that actively creates scenarios of maximum suffering.
Existential · Group 18 • Noble
Scenario where humanity becomes economically, scientifically, and strategically irrelevant in a world dominated by superintelligent AI, even without active hostility.
Use the category hubs for a fast overview, then jump straight into representative risk pages before opening the full interactive table below.
14 risks
Impacts on jobs, markets, economic concentration, and value distribution.
3 risks
Energy and material externalities across the AI model lifecycle.
22 risks
Long-horizon alignment, loss-of-control, and catastrophic-risk scenarios.
8 risks
Interaction risks in dependence, overtrust, perception, and AI-assisted decision-making.
6 risks
Regulatory, compliance, rights, and liability exposure.
10 risks
Intentional misuse vectors including fraud, manipulation, and model weaponization.
8 risks
Risks related to exposure, leakage, surveillance, and re-identification of data.
11 risks
Failure modes that degrade output quality, consistency, or trustworthiness.
9 risks
Technical and cybersecurity attack vectors affecting model integrity, control, and resilience.
27 risks
Collective and system-level impacts such as bias, inequality, information harm, and social damage.