A risk and safety hub for understanding how AI systems fail, how harms emerge, and how controls reduce unsafe behavior.
30
Terms
7
Comparisons
8
Questions
AI safety and risk focuses on preventing, detecting, and responding to failures that can harm people, organizations, or public trust. It covers technical reliability, organizational controls, foreseeable misuse, monitoring, and incident escalation.
Safety failures often appear after deployment, when real users, changing data, and unexpected contexts expose weaknesses that were not visible in a benchmark.
Useful for safety cases, AI risk assessments, incident playbooks, post-market monitoring, healthcare and public-sector AI reviews, and high-risk deployment governance.
Core concepts for distinguishing safety, ethics, risk, harm, and alignment problems.
Terms describing how systems degrade, become unreliable, or behave differently after deployment.
Methods used to test, monitor, and respond to safety issues across the AI lifecycle.
A side-by-side comparison of [AI] Safety and [AI] Ethics. Understand how harm prevention practices relate to broader questions of human values, rights, and responsible design.
A side-by-side comparison of Risk and Harm. It explains how a likelihood-and-severity assessment differs from the adverse effect experienced by a person, group, organization, or system.
A side-by-side comparison of Data Drift and Model Drift. Understand how changes in input-data distribution differ from changes or degradation in model performance over time.
A side-by-side comparison of Data Drift and Concept Drift. Understand whether a model problem comes from changing input distributions or changing relationships between inputs and outcomes.
A side-by-side comparison of Model Drift and Concept Drift. It explains how performance degradation over time differs from a changing statistical relationship between inputs and target outcomes.
A side-by-side comparison of Safety Case and AI Risk Assessment. Understand how the concepts differ, when each term applies, and why the distinction matters for AI governance, evaluation, or system design.
Distinguish between Prompt Injection (hijacking model behavior via malicious input) and Jailbreaking (bypassing safety guardrails to produce restricted content). Both are LLM security threats but differ in mechanism and goal.
Use Data Drift when input data distribution changes; use Concept Drift when the relationship between inputs and the target outcome changes.
Alignment is the practice of designing, training, and governing an AI system so that its goals, outputs, and behavior remain consistent with human intentions, values, and acceptable use constraints.
An AI Incident is an event or series of events in which the development, deployment, use, or malfunction of an AI system causes or could cause harm. Such harm may involve safety, discrimination, privacy, legal compliance, critical infrastructure, property, communities, or the environment.
HITL, or human-in-the-loop, is an approach in which people remain involved in shaping, reviewing, approving, or improving AI system outputs or decisions. It is used to combine machine efficiency with human judgment, especially in higher-risk or context-sensitive tasks.
AI Toxicity is harmful, offensive, false, biased, or otherwise inappropriate output generated or amplified by an AI system. It may arise from training data, system design, prompt context, misalignment, or inadequate safeguards.
A post-market monitoring system consists of activities by AI system providers to collect and review experience from systems placed on the market or put into service. Its purpose is to identify whether corrective or preventive action is needed after deployment.
Anomaly detection is the process of identifying observations, events, or patterns that differ significantly from expected behavior. In AI and data analysis, it is used to flag outliers that may indicate errors, fraud, security incidents, system faults, or other unusual conditions.
Security by design is an approach in which security requirements and safeguards are integrated from the start of system development. For AI systems, it includes anticipating misuse, protecting data and models, and embedding access controls, monitoring, and resilience throughout the lifecycle.