ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›AI MEANINGS
AI MEANINGS

AI Glossary & Meanings

The AI glossary in plain English: every AI word with a one-line answer, A to Z, and a full page with an everyday comparison and example for the big ones.

PAGES131
WORDS145,201
SOURCES1,307
CHECKED1 OCT 26

Heard an AI word and not sure what it means? Search for it below, or pick a topic. New to AI altogether? Start with AI explained for complete beginners, or print the one-page cheat sheet.

157 words, sorted by topic. Search, or pick a topic. Cards marked FULL MEANING open a full plain-English page (141 so far); the rest give a quick definition.

SHOWING 157 OF 157 WORDS

Start here

17 WORDS

The words everyone meets first.

Artificial intelligence (AI)
Computer software that can do tasks we normally think need a human mind, such as understanding language, recognising pictures, deciding or writing.
FULL MEANING →
Machine learning
Teaching computers by showing them many examples, so they find the patterns themselves instead of following rules a person wrote.
FULL MEANING →
Deep learning
A kind of machine learning that passes information through many stacked layers of a neural network; it is the method behind chatbots and image makers.
FULL MEANING →
Neural network
A program built from many small calculating units joined in layers, which learns from examples by adjusting the strength of the connections between them.
FULL MEANING →
Chatbot / AI chatbot
A program you talk to in ordinary words, by typing or speaking; an AI chatbot writes each reply fresh instead of following a fixed script.
FULL MEANING →
AI model
The trained part of an AI tool: a saved set of patterns learned from data, which the app you use relies on to answer, predict or create.
FULL MEANING →
LLM (large language model)
The kind of AI that reads and writes text: the engine inside chatbots, which learned the patterns of language from a huge amount of writing.
FULL MEANING →
Generative AI
AI that creates new content, such as text, pictures, video, music or voices, from a description of what you want.
FULL MEANING →
Algorithm (in AI)
A set of step-by-step instructions a computer follows to reach a result; in AI it is the method that trains a model, and in apps it picks what you see.
FULL MEANING →
Natural language processing (NLP)
The part of AI that helps computers understand and use everyday human language, so they can follow spoken requests, answer questions and write replies.
FULL MEANING →
Prompt
The message you give an AI tool: the question, request or instructions it works from. The prompt is the input; the answer is the output.
FULL MEANING →
Token
The unit models actually read — roughly three-quarters of an English word. Context windows, prices, and speeds are all measured in these.
FULL MEANING →
Hallucination
When an AI gives a confident, fluent answer that is wrong or made up. It is not lying; it produces what sounds likely, so check anything that matters.
FULL MEANING →
Agent
A model given tools, memory, and a goal, allowed to take multiple steps on its own. The word covers everything from a script that retries twice to software that plans a workday.
FULL MEANING →
Expert system
An early kind of AI, first built in the 1970s, that copies how a human expert makes decisions in one narrow area by applying “if this, then that” rules to a store of facts written in by experts, instead of learning from data.
QUICK DEFINITION
GPT
Short for Generative Pre-trained Transformer: OpenAI’s family of large language models. ChatGPT is the chat app that runs on them.
FULL MEANING →
Narrow AI (weak AI)
AI built for one job or a small set of jobs, such as recognising faces or playing chess; it can be superhuman at that job and useless at anything else.
FULL MEANING →

Using AI

30 WORDS

Getting better answers from the tools you use.

AI literacy
Knowing enough about AI to use it well and safely: what it can and cannot do, how to check its answers, what happens to your data and when not to use it.
FULL MEANING →
AI memory
A chatbot feature that saves details about you from past chats and reuses them later; it is a setting you can usually view, pause or delete.
FULL MEANING →
AI prompt generator
A tool that turns a short description of what you want into a longer, more detailed prompt; a useful first draft that you still need to edit and test.
FULL MEANING →
AI search engine
A search tool that answers in written sentences, built from web pages it looks up as you ask, with links to those pages, instead of only giving a list of links. Perplexity and Google’s AI Mode are examples.
FULL MEANING →
Chain of thought
A model working through intermediate reasoning before answering. On newer reasoning-trained models this happens internally — asking for it explicitly is often redundant.
FULL MEANING →
Context engineering
Choosing and arranging everything an AI model sees while it works (instructions, documents, examples, tools, memory, earlier messages) so it has what it needs and nothing that crowds it out. The next step on from prompt engineering.
FULL MEANING →
Context rot
An AI getting less accurate as it is given more and more text at once, so long chats and huge document piles make it miss details.
FULL MEANING →
Context window
The maximum amount of text, in tokens, a model can consider at once. Has grown from a few thousand tokens in early models to far larger windows; check the current figure for the model you use. Ordering within it now matters more than fitting.
FULL MEANING →
Contextual prompting
Giving an AI the background it needs (who it is for, what it is for, what good looks like) instead of hunting for magic wording.
FULL ARTICLE →
Deep research
An AI feature that turns one question into a research task: it makes a plan, runs many web searches, reads the pages and, after several minutes, writes a long report with links to its sources.
FULL MEANING →
Few-shot prompting
Asking the AI with no examples (zero-shot) or with a few examples of what you want (few-shot). A couple of good examples often help a lot.
FULL MEANING →
Grounding
Tying a model answer to supplied or retrieved sources instead of its trained memory. The main defense against fabricated detail.
FULL MEANING →
In-context learning
When an AI picks up how to do a task from examples written into the prompt, without any retraining: its internal settings (parameters) stay exactly as they were.
QUICK DEFINITION
Knowledge cutoff
The date an AI model’s training data stops; it knows nothing after it unless the app lets it search the web or read new documents.
FULL MEANING →
Latency
How long the AI takes to start and finish answering. Running a model on your own device avoids the trip over the internet.
FULL MEANING →
Machine translation
A computer translating text or speech from one language into another, as translation apps do; good for the gist, but it can drop words, garble names and miss the tone.
FULL MEANING →
Multimodal
An AI that can take in or produce more than text: pictures, sound, video or documents.
FULL MEANING →
Prompt caching
Reusing the unchanged start of a long prompt so the AI does not reprocess it every time, which makes repeat requests faster and cheaper.
FULL ARTICLE →
Prompt chaining
Splitting a big job into a series of smaller prompts, where each answer feeds the next step, so you can check the work between stages.
FULL MEANING →
Prompt engineering
Writing and testing the instructions you give an AI so it reliably produces what you need: clear goals, the right background, and checking the results.
FULL MEANING →
Reasoning model
A model trained to deliberate internally before answering — slower, costlier, stronger on hard problems. Brief it with goals and budgets, not step-by-step scripts.
FULL MEANING →
Recommendation algorithm
The system an app uses to choose what to show you next, such as the videos in your feed, guessed from what you and people like you watched, liked or skipped.
FULL MEANING →
Role prompting
Telling an AI who to be before you ask, such as “You are a patient maths tutor”. It mainly changes the tone, focus and level of the answer, not how accurate it is.
FULL MEANING →
Sentiment analysis
Using AI to work out the feeling behind a piece of text, such as whether a review is positive, negative or neutral. Also called opinion mining; sarcasm and mixed feelings are its weak spots.
FULL MEANING →
Skill (AI skill)
A saved folder of written instructions that an AI assistant reads only when a task calls for it, so it does that job the same way every time.
FULL ARTICLE →
Streaming
Seeing the AI’s answer appear word by word as it is written, instead of waiting for the whole thing.
QUICK DEFINITION
Sycophancy
A chatbot telling you what you want to hear: agreeing with you, praising your ideas or backing down when you push, even when the facts say otherwise.
FULL MEANING →
System prompt
The standing instructions an AI receives before your message, setting its role, rules and limits.
FULL MEANING →
Temperature
The randomness dial. Low means consistent and repetitive; high means varied and risky. Check the vendor’s advice before changing it: on Gemini 3.x Google recommends leaving it at the default (sourced here).
FULL MEANING →
Zero-shot prompting
Asking the AI to do a task with instructions only, without showing it any examples of the answer you want. Most everyday prompts are zero-shot.
FULL MEANING →

Building & agents

22 WORDS

How AI is wired into apps and given jobs to do.

Agentic AI
AI that works towards a goal on its own, planning and taking several steps with little supervision, instead of giving one answer and stopping.
FULL MEANING →
Agentic workflow
A way of using AI in several steps instead of one: the model plans, uses tools, checks and revises its work before you get the result. The steps may be set by a person or chosen by the AI.
FULL MEANING →
API
The way software talks to an AI without a chat window: a program sends a request with the prompt and settings, and gets the answer back.
FULL MEANING →
Batch processing
Sending many requests together to be answered later rather than instantly; providers usually make it cheaper for jobs that can wait.
QUICK DEFINITION
Benchmark
A standardized test set used to compare models. Useful directionally; notoriously gameable; never a substitute for testing on your own cases.
FULL MEANING →
Coding agent
An AI tool that does programming work itself: it reads a project’s files, makes changes, runs commands such as tests and keeps going until the task is done. Claude Code, Codex and Gemini CLI are examples.
FULL MEANING →
Embedding
A list of numbers representing meaning. Texts with similar meaning get nearby numbers, which is what makes semantic search and RAG possible.
FULL MEANING →
Eval
A repeatable test of an AI system: give it set inputs, then grade the outputs against your own standards. The difference between “it seems fine” and knowing.
FULL MEANING →
Function calling
When an AI replies with a structured request for a program to run (search, look something up, calculate) instead of just writing text; your app runs it and hands back the result. It is what lets agents act.
FULL MEANING →
Human-in-the-loop
A setup where a person must check and approve the AI’s work before it is sent, published or acted on.
FULL MEANING →
MCP (Model Context Protocol)
A shared standard that lets AI assistants connect to other apps, files and tools, so a chatbot can work inside them instead of only chatting.
FULL MEANING →
Multi-agent system
A setup where several AI agents work on one job together, often a lead agent that splits the work, hands it to helper agents and combines what they find.
FULL MEANING →
Orchestrator
The part of an AI system, usually a lead AI but sometimes a person or plain code, that splits a big goal into tasks, hands them to workers and combines the results.
FULL MEANING →
Plugin
An add-on package you install into an AI app to give it new abilities in one go. In tools like Claude, ChatGPT and Codex it bundles skills, connectors and sometimes helper agents.
FULL MEANING →
RAG
Retrieval-Augmented Generation: search first, then have the model answer from what was found. The standard pattern for grounding answers in your documents.
FULL MEANING →
Rate limit
A cap, set by the provider, on how many requests or tokens you can send to an AI service in a set time. Go over it and requests are refused for a while, which is why apps queue, retry and batch.
FULL MEANING →
Schema / structured output
Making an AI answer in an exact, machine-readable shape, usually JSON that follows a template (schema) you supply, so another program can use the result directly. It fixes the shape, not the facts.
FULL MEANING →
Subagent
A helper AI that a main AI agent sends off to do one part of a job in its own separate workspace, then report back a short summary.
FULL MEANING →
Tool use
When an AI model asks another program to do a job, such as a web search, so it can fetch facts or act. Function calling is the developer name for the same idea.
FULL MEANING →
Vector database
A database built to store embeddings and find nearest-meaning matches fast. The retrieval half of RAG at scale.
FULL MEANING →
Vibe coding
Making software by describing what you want to an AI in plain words and letting it write the code, often without reading the code yourself.
FULL MEANING →

How AI is made

48 WORDS

Training, models and the ideas behind them.

AGI (artificial general intelligence)
An AI that could do most mental work as well as a person. There is no agreed definition, which is why claims about it are hard to check.
FULL ARTICLE →
AI bias
When an AI system works better for some people than others or repeats unfair stereotypes, usually because of the data and human choices behind it.
FULL MEANING →
AI winter
A period when disappointment with AI makes funding and interest dry up after a boom of big promises. There have been two: in the 1970s, and from the late 1980s into the 1990s.
FULL MEANING →
Alignment
The work of making an AI behave the way people intend: helpful, honest and refusing to cause harm.
FULL MEANING →
Attention
The part of a transformer that weighs how much each word, or part of the input, matters to every other part. It is why the surrounding text changes the answer.
FULL MEANING →
Backpropagation
How a neural network learns from its mistakes: after it makes guesses and measures how wrong they were, it works backwards through its layers to decide which connections to strengthen and which to weaken. It is how gradient descent is carried out in neural networks.
QUICK DEFINITION
Base model
A model that has finished its first big round of training but has not yet been tuned for a particular job. It is the starting point that fine-tuning builds on.
QUICK DEFINITION
Checkpoint
A saved snapshot of a model at a point in training. What you actually download when you download a model.
FULL MEANING →
Classification and regression
Two kinds of prediction. Classification picks a category (is this email spam? which language is this sentence in?); regression predicts a number (what will this house sell for? how much rain will fall?).
QUICK DEFINITION
Clustering
Getting an AI to sort examples into groups of similar items, usually from data that has no labels; a person can then decide what each group means.
FULL MEANING →
Computer vision
The part of AI that lets computers make sense of pictures and video, such as spotting objects, faces, text or movement.
FULL MEANING →
Constitutional AI
A way of training an AI with a written list of principles (a “constitution”) that the AI uses to critique and improve its own answers, so fewer human ratings are needed. Anthropic developed it for its Claude models.
FULL MEANING →
Dataset
A collection of data gathered together, often laid out like a spreadsheet. In AI, part of a dataset is used to train a model and other parts are kept back to check it.
QUICK DEFINITION
Distillation
Training a smaller model to copy a larger one, so it is cheaper and faster to run, usually with some loss in ability.
FULL MEANING →
Emergent abilities
Skills that larger AI models seem to gain suddenly while smaller ones lack them; researchers still dispute whether the jumps are real or a measuring effect.
FULL MEANING →
Epoch
One full pass of training through the whole set of training examples, in which the model has worked through every example once.
QUICK DEFINITION
Fine-tuning
Continuing a model training on your own examples so it adopts a style, format, or domain. Heavier than prompting; lighter than pretraining.
FULL MEANING →
Foundation model
A big, general AI model trained on a huge amount of varied data, used as the starting point for many different tools and jobs.
FULL MEANING →
Frontier model
The most capable AI models available at a given moment, the ones every other model is compared against.
FULL MEANING →
Gradient descent
The step-by-step method used to shrink the loss during training: it nudges the model’s internal numbers (its weights) a little at a time in the direction that makes the errors smaller, again and again.
QUICK DEFINITION
Inference
Using a trained model to get an answer, as opposed to training it. Every question you ask a chatbot is inference.
FULL MEANING →
Instruction tuning
Extra training (a kind of fine-tuning) on many example instructions covering lots of different tasks, so the model gets better at doing what it is asked, even for requests that come with no worked example.
QUICK DEFINITION
Labelled data
Examples that come with the right answer attached, such as emails each marked “spam” or “not spam”. A model learns from labelled examples, then makes guesses about new ones that have no label.
QUICK DEFINITION
Latent space
The squeezed-down form a model turns data into on the inside: each item is boiled down to a compact set of numbers that keeps only its most important features, and similar items end up close together.
FULL MEANING →
Loss function
The scoring rule used while an AI is trained: it works out how far the model’s guesses are from the right answers, giving a lower score (the “loss”) for better guesses. Training tries to make that score as small as possible.
QUICK DEFINITION
Mixture of Experts (MoE)
A design where a big model is split into specialist parts and only a few are used for each request, so it answers faster; the whole model still has to fit in memory.
FULL MEANING →
Model collapse
AI getting worse when it is trained mostly on AI-made content instead of human-made data. Studies differ on how far keeping real data in the mix prevents it.
FULL MEANING →
Open vs closed models
Open models publish their files so anyone can run them; closed models are used only through the maker’s app or service.
FULL ARTICLE →
Open weights
A model whose trained parameters are downloadable. Licenses vary from fully permissive to research-only; the license, not the adjective open, is what matters.
FULL MEANING →
Overfitting
When an AI learns its practice examples so exactly that it does well on them but badly on anything new, like memorising answers instead of the subject.
FULL MEANING →
Parameters (weights)
The numbers inside an AI model that it adjusts for itself while learning; together they hold everything it picked up in training.
FULL MEANING →
Pretraining
The first and biggest stage of training: the model learns language and facts about the world by predicting the next token across a vast amount of text, before later training teaches it to follow instructions.
FULL MEANING →
Reinforcement learning
Teaching an AI by trial and error: it tries actions, gets a reward or a penalty for the result, and learns which actions earn the most reward.
FULL MEANING →
RLHF
Reinforcement learning from human feedback: people rate the AI’s answers, and the model is trained towards the ones they preferred.
FULL MEANING →
Scaling laws
The pattern researchers measured: models made bigger, with more data and computing power, make fewer prediction errors. Whether that keeps turning into more capability is debated.
FULL MEANING →
Singularity (technological singularity)
The predicted moment when AI becomes smarter than people and improves itself so fast that the future can no longer be forecast. A prediction, not an event, and strongly disputed.
FULL MEANING →
Superintelligence (ASI)
A hypothetical AI far smarter than the best humans at almost everything. Some AI companies say they are building it; whether and when it could exist is unknown.
FULL MEANING →
Supervised learning
Teaching an AI with examples that already carry the right answer, so it learns to give good answers for new examples on its own.
FULL MEANING →
Synthetic data
Training data made by computers rather than collected from people. Useful when used carefully; risky when AI-made content quietly replaces real data.
FULL MEANING →
Training data
The examples an AI studies before anyone uses it, such as text or pictures, whose patterns shape everything it can later say or do.
FULL MEANING →
Training run
One full session of training a model on a large amount of data, using many specialised computer chips: the long and costly part of making an AI.
FULL ARTICLE →
Transfer learning
Reusing what a model learned on one task to help it with another, such as carrying knowledge from a task with plenty of example data over to one with very little.
QUICK DEFINITION
Transformer
The 2017 attention-based architecture underlying most modern language models and much else. The frame every floor is bolted to.
FULL MEANING →
Turing test
Alan Turing’s 1950 “imitation game”: can a machine chat so much like a person that a judge cannot tell which is which? Passing shows it sounds human, not that it understands.
FULL MEANING →
Unsupervised learning
Teaching an AI with data that has no answers attached, so it finds patterns and groups by itself; a person then decides what the groups mean.
FULL MEANING →
Vision-language model (VLM)
An AI model that takes in pictures (and often video) as well as text and answers in words, such as describing a photo, reading a document or answering questions about what it sees.
FULL MEANING →
Weights
The learned numbers that make up a trained model. Apps and chat windows are just ways of reaching them.
FULL MEANING →
World model
An AI system that has learned how a place or situation behaves well enough to simulate it and predict what happens next. Used to train robots and agents, and to generate worlds people can explore.
FULL MEANING →

Hardware & running it yourself

7 WORDS

Chips, memory and AI on your own device.

Edge AI
Running AI directly on a local device, such as a camera, sensor, smartwatch or car, instead of sending the data to a distant server, so answers come back faster.
QUICK DEFINITION
GPU
The parallel processor that trains and often runs models. Its memory, VRAM, is usually the binding constraint for running models locally.
FULL MEANING →
Local LLM / on-device AI
A model running entirely on your own computer or phone: your words never leave the device, and what it can run is limited by your memory.
FULL MEANING →
NPU
Neural processing unit — the accelerator in modern phones and laptops built specifically to run models efficiently on-device.
FULL MEANING →
Quantization
Storing a model’s numbers with less precision so it needs less memory and often runs faster, at a small cost in quality. It is how big models fit on laptops.
FULL MEANING →
Small language model (SLM)
A language model built much smaller than the big chatbot models, so it can run on a phone or laptop, even offline, at the cost of knowing less.
FULL MEANING →
VRAM
The memory on a graphics card. It decides how big a model your computer can run; quantization is the usual way to fit a bigger one.
FULL MEANING →

Images, video & voice

18 WORDS

The words creators meet in image, video and voice tools.

AI upscaling
Using AI to make a small or blurry image bigger and sharper by filling in new detail it predicts, rather than just stretching the pixels.
FULL MEANING →
CFG scale (guidance scale)
Classifier-free guidance scale, also called guidance scale: the image-tool setting for how strictly the picture follows your prompt. Too high can cause visible flaws; the right value depends on the model.
FULL MEANING →
ControlNet
An add-on for AI image models that makes a picture follow a guide image’s structure, such as a body pose, outlines or depth, while your prompt decides how it looks.
FULL MEANING →
Diffusion model
The architecture behind many AI image and video generators: it learns to turn noise into pictures step by step, steered by your description.
FULL MEANING →
GAN (generative adversarial network)
A way of making new data, such as realistic faces of people who do not exist, using two parts that compete: a generator makes new examples, and a discriminator tries to tell them apart from real data. The generator tries to fool the discriminator; the discriminator tries not to be fooled.
QUICK DEFINITION
Image-to-video
Giving AI a still picture and having it turn that picture into a short moving clip, with your words saying what should move.
FULL MEANING →
Img2img (image-to-image)
Short for image-to-image: giving an AI image model a picture plus a written prompt, so it makes a new picture that starts from yours. A strength setting decides how much changes.
FULL MEANING →
Inpainting and outpainting
Inpainting is AI repainting just one marked part of an image; outpainting is AI extending a picture beyond its edges with new matching scenery.
FULL MEANING →
Keyframe
A frame where you fix how something looks or where it is at a set moment, so the software fills in the frames in between. In AI video, usually the start and end images you give the model.
FULL MEANING →
Lip sync
Making a face’s mouth movements match a voice track, so a person, character or photo looks as if it is saying the words. AI tools either edit the mouth to fit audio or generate speech and mouth together.
FULL MEANING →
LoRA
Low-Rank Adaptation: small trainable add-on weights that customize a model cheaply — the standard way styles and subjects are added to open image models.
FULL MEANING →
Negative prompt
A separate list of things you do not want an AI image or video tool to show, though not every tool reads one.
FULL MEANING →
Seed (in AI images)
The number that sets the random starting point of an AI image, so reusing it with the same prompt and settings usually repeats the picture.
FULL MEANING →
Speech-to-text (speech recognition)
Software that listens to spoken words and writes them down as text, as in phone dictation, automatic captions and meeting transcripts. The opposite of text-to-speech.
FULL MEANING →
Style reference
A picture you give an AI image tool so new images copy its look (colours, textures, lighting, feel) without copying what is in it. Midjourney’s --sref is one example.
FULL MEANING →
Text-to-image
Typing a description and having AI make a brand-new picture to match it, instead of drawing or photographing it yourself.
FULL MEANING →
Text-to-speech (TTS)
A computer reading written words out loud in a voice that sounds human, used for audiobooks, voice assistants and spoken announcements.
FULL MEANING →
Text-to-video
Typing a description and having AI make a short video clip to match it, with no camera, actors or filming involved.
FULL MEANING →

Safety & trust

15 WORDS

Fakes, attacks and how to tell what is real.

AI ethics
The principles for building and using AI without harming people: fairness, privacy, openness about how it works, safety, and a person answerable when it goes wrong. Companies often call it responsible AI.
FULL MEANING →
AI governance
The rules, standards and oversight that decide how AI is built and used: laws such as the EU AI Act, voluntary frameworks and standards, and an organisation’s own policies and checks.
FULL MEANING →
Dead internet theory
The claim that much of the web is now made and read by bots rather than people. Parts of it are measurable; much of it is not.
FULL ARTICLE →
Deepfake
A fake video, photo or voice recording, made with AI, that shows a real person doing or saying something they never did.
FULL MEANING →
Explainable AI (XAI)
AI that can give an understandable reason for its answer, instead of being a “black box” whose workings even its makers cannot fully see.
FULL MEANING →
Guardrails
The rules, filters and checks that keep an AI within bounds: some sit around the model, checking what goes in and out; some are built into its training.
FULL MEANING →
Jailbreak
A prompt written to trick an AI into ignoring its safety rules.
FULL MEANING →
Misinformation
False or misleading information shared as if it were true, whether by mistake or on purpose. AI makes it cheaper to produce.
FULL ARTICLE →
Prompt injection
Hidden instructions inside something an AI reads, such as a web page or email, that try to hijack what it does. The classic attack on AI agents.
FULL MEANING →
Provenance / C2PA
Information attached to a photo, video or audio file that records how it was made and edited, so you can check whether it is what it claims to be.
FULL MEANING →
Red-teaming
Systematically attacking your own model or app — jailbreaks, injections, edge cases — before someone else does.
FULL MEANING →
Shadow AI
Staff using AI tools at work without their employer’s approval, such as pasting work documents into a personal chatbot account. It can leak private data and break company or legal rules.
FULL MEANING →
Slop
Low-effort AI-made content published in bulk for clicks and ad money. The problem is the flood, not AI itself.
FULL MEANING →
Voice cloning
Using AI to copy a real person’s voice from a recording so it can say new words; honest uses need that person’s consent, and scammers use it to fake calls.
FULL MEANING →
Watermarking
Embedding a detectable signature in generated output to mark it as synthetic. Technically fragile, politically inevitable, actively evolving. Claude models in use today, including Opus 5 and Opus 5.5, already watermark their text to comply with the EU AI Act; Anthropic’s help page now ticks text watermarks for nine Claude models, from Fable 5.1 to Haiku 5.5; older ones such as Opus 4.5, Sonnet 4.5 and Haiku 4.5 are ticked only for file credentials so far. Only organisations granted access can check for the mark for now, and a detected mark is a signal, not proof. More on Claude’s text watermark.Claude Help Center, How Claude marks AI-generated content, read at source 13 Oct 2026: “Generated text will carry embedded watermarks”; “Watermark detection is currently in private preview, available to eligible organizations as required under EU law”, naming regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations and EU civil society groups; “A detected mark provides a signal that content was processed by Claude, but is not fully conclusive.” Its model table “Which Claude models support watermarking”, read the same day, ticks “Text watermarks in Claude output” for Fable 5.1, Fable 5, Mythos 5.1, Opus 5.5, Opus 5, Opus 4.8, Sonnet 5.5, Sonnet 5 and Haiku 5.5, and only Content Credentials (C2PA) for Mythos 5, Opus 4.7, 4.6 and 4.5, Sonnet 4.6 and 4.5 and Haiku 4.5, and the page says “Anthropic is adding watermarks to outputs from models released before August 2, 2026.” On 23 Sep 2026 the table ticked only Fable 5.1, Mythos 5.1, Opus 5.5 and Opus 5, and that sentence ended “with all covered by December 2, 2026”; the date is no longer on the page.Anthropic, Introducing Claude Opus 5.5, 22 Sep 2026, read at source 23 Sep 2026: “As with Fable 5.1, Opus 5.5 comes with our watermarking measures to comply with the EU AI Act”.Anthropic, How Claude’s text watermark works, 14 Aug 2026, read at source 22 Sep 2026: “Future Claude models will generate text that contains a watermark. This is a way of determining the likelihood that Claude was involved in writing the text, and we, along with several other major AI providers, are implementing this change to comply with the EU AI Act.”First-hand: until 23 Sep 2026 this entry said Anthropic’s post “describes future models, not current ones”. That was wrong when written: Anthropic, Claude Fable and Mythos 5.1, September 2026, read at source 22 Sep 2026: “This required us to add a watermark” … “to the outputs of models released after August 2, 2026.”
FULL MEANING →

Related on this site: AI Flashcards

aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE