Plain-language definitions of the terms security, risk and compliance teams meet when they put AI to work, each traced to the regulation, standard or research behind it.
Every term sits in one of five topics. Pick one to filter the index, or search across all of them.
17 terms
Frameworks, assessments, inventories and the oversight that keeps AI accountable.
See the terms →
9 terms
The AI your vendors build, embed and run, and the evidence to ask for.
See the terms →
15 terms
The EU AI Act, NIST AI RMF, ISO/IEC 42001 and the terms they define.
See the terms →
12 terms
Models, datasets and packages: where they come from and how they get compromised.
See the terms →
11 terms
Attacks on models and agents, and the controls that stop them.
See the terms →Each entry opens with a short answer you can quote, then the detail: what it covers, how it works and what the rules say.
64 terms
Securing AI agents that plan, call tools and act, beyond a single prompt and response.
→The rules that tell staff which AI tools they may use, for what and with which data.
→Measuring, evaluating and communicating whether an AI system can be trusted.
→An independent review of an AI system or AI management against defined criteria.
→An inventory of the models, data, code and infrastructure inside an AI system.
→Cranium AI’s living, exportable transparency report for an AI system.
→Dependence on a few providers for the models, cloud and chips behind your AI.
→The terms buyers add to vendor contracts to control training use, model changes and exit.
→The policies, roles and controls that keep an organization’s AI lawful, safe and accountable.
→The policies, roles and controls that decide which AI you build, buy and run.
→Software that keeps the record of your AI systems and runs governance against it.
→The controls that keep an AI system’s inputs, outputs and actions inside limits its owner has set.
→A documented evaluation of how an AI system could affect people and society.
→An event where an AI system leads to harm, and what the EU AI Act says to do next.
→A maintained record of every AI system you build, buy or use, with owner and risk.
→The skills and understanding people need to use AI responsibly, required by Article 4.
→The policies, roles and processes an organization uses to direct and control its AI.
→The EU AI Act’s split between who builds and markets an AI system and who uses it.
→Adversarial testing that attacks an AI system the way a real adversary would, before one does.
→A structured evaluation of what could go wrong with an AI system, and how badly.
→The continuous practice of identifying, assessing, treating and monitoring AI risk.
→Protecting AI systems while they run, by watching live prompts, responses and agent activity.
→Finding every AI system, checking how it is built and exposed, and fixing the gaps.
→A company your vendor engages to process your data for an AI feature, such as a model provider.
→Controlling the risks carried by the models, data, code and vendors behind your AI.
→A public or gated page of evidence on how an organization builds, governs and secures AI.
→The request that registers, tiers and routes a proposed AI use before it starts.
→A structured review of the AI inside a supplier’s product, ending in a tier and a decision.
→MITRE’s public knowledge base of the tactics and techniques adversaries use against AI systems.
→The CycloneDX bill of materials for machine learning models and the datasets behind them.
→Hidden behavior in a model that stays dormant until an attacker’s trigger appears in the input.
→A short document stating what a model is for, how it was evaluated and where it falls short.
→The open-source standard that connects AI applications to external tools, data sources and workflows.
→The verifiable record of where an AI model came from and what has changed since.
→How banks govern, validate and monitor the models behind their decisions.
→Static inspection of model files for malicious code and unsafe serialization before they are loaded.
→Code hidden in a saved model file that executes the moment someone loads the model.
→The independent check that a model performs as expected, before and after it goes live.
→When a coding model invents a package name that an attacker can register and fill with malicious code.
→How providers track a high-risk AI system’s performance after it is placed on the market.
→The uses of AI that Article 5 of the EU AI Act bans outright, with the highest fines.
→Input that makes a language model act against its developer’s intent, typed by a user or hidden in content.
→Try a shorter word or another topic, or clear the filters to see every term.
How the two disciplines differ, and why enterprise AI needs them to converge into one loop.
Read the guide →
The hub for vendor AI risk: straight answers, the AI Card workflow and a library to choose, assess, monitor and prove.
Read the guide →
The AI bill of materials explained: what goes in one, how it differs from an SBOM and how to build one.
Read the guide →See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.
Cookie Policy · Privacy Policy
Global Privacy Control detected. Opt-out request honored: marketing tags stay off in this browser.