The user wants to understand Artificial Intelligence in the context of Generative AI and apply it to practical AI governance or compliance work.
Use Artificial Intelligence for the broad field and system category; use Generative AI when the system creates new content from prompts or context.
Artificial intelligence is the broad field and system category covering machines that perform tasks associated with human intelligence, such as prediction, classification, perception, planning, optimisation, language processing, or decision support. Generative AI is a narrower category of AI focused on creating new content from prompts or context. In Caesar AI Atlas, the key terms are artificial-intelligence, generative-ai, and ai-system. An AI system may classify credit risk, detect fraud, route a delivery vehicle, recommend a product, recognise an image, or generate a document. Only some of those systems are generative. Generative AI includes tools that produce text, code, images, audio, video, synthetic data, or designs. The distinction matters because governance, user expectations, and controls differ. A prediction model needs evaluation for accuracy and fairness; a generative model also needs controls for hallucination, provenance, synthetic media, prompt injection, and output misuse.
Artificial intelligence is the whole toolbox. Generative AI is one type of tool inside that toolbox. A fraud detection model, a spam filter, a route optimiser, and a face recognition system can all be AI without being generative AI. A chatbot that drafts an email, an image generator that creates a poster, or a coding assistant that writes a function are generative AI because they create new content. The terms overlap, but they should not be used as synonyms.
Analogy
AI is like the category 'vehicles'; generative AI is like 'delivery vans' — important, common, and useful, but not the whole category.
The distinction matters because teams often write one policy for all AI and then miss risks specific to generative systems. Traditional AI governance may focus on model performance, bias, explainability, data quality, and decision impact. Generative AI governance must also address content provenance, hallucination, confidential input leakage, copyright, misinformation, impersonation, deepfakes, and prompt-based attacks. Under the EU AI Act, general-purpose AI and transparency duties create additional implementation pressure, with GPAI provider obligations applying from 2 August 2025 and broader obligations phasing in later. Organisations should classify systems accurately before choosing controls.
Urgency
Do not wait until rollout: classify whether a system is generative during intake, procurement, or architecture review.
A reliable AI intake process should first identify whether the tool is an AI system and then whether it is generative AI, general-purpose AI, high-impact decision support, or another category. The classification should drive testing, notices, logging, human review, security controls, and vendor documentation requests. For generative AI, teams should add content-specific controls, including output review, source grounding, watermarking or labeling where required, and restrictions on sensitive prompts.
The most common mistake is calling every AI tool 'generative AI' because chatbots dominate public discussion. The opposite mistake is treating generative AI like a normal prediction model and missing content-specific risks.
Mistake 1: Applying only accuracy metrics to a generative assistant can miss hallucinations, unsupported claims, and unsafe generated content.
Mistake 2: Using a generic AI policy without prompt, output, and disclosure rules can leave employees unsure how to handle generated material.
This page should connect the broad artificial intelligence concept to generative AI and AI system definitions. It also links to comparisons between generative AI and artificial intelligence, and to pages on large language models and diffusion models.