Copyright, deepfakes and what is real
Two questions come up the moment you make or see AI content: who owns it, and is it real. Rules vary by country and are changing, so treat this as the basics.
Words on this page (5)
- training
- : Adjusting a model's parameters on data so its predictions improve.
- tool
- : A function the model can ask your program to run, described by a name, description and argument schema.
- deepfake
- : A realistic fake image, video or voice made with AI to look like a real person or event.
- C2PA
- : A standard for attaching a signed record of where an image or video came from and how it was edited.
- trace
- : The full record of one run: inputs, model calls, tool calls, outputs, timing and cost.
Who owns what an AI makes
In the United States, the Copyright Office has said that copyright needs human authorship. Material produced purely by an AI is not protected, but your own writing, your selection and arrangement, and your edits can be. Other countries differ. Whether models on copyrighted work is lawful is being decided in courts right now, with several large cases still open. For you, the practical rules are simple: do not claim AI output as hand-made work where it matters, do not present an AI image as a real photograph, and check the terms of the for commercial use.
Deepfakes
A is fake audio, video or imagery that looks real. A few seconds of a person's voice can be cloned well enough to fool relatives. Common scams use a cloned voice to say "I'm in trouble, send money now". Defenses: hang up and call back on a number you already know, agree a family code word, and be slow when someone creates urgency. For images and video, look at the source before the picture. Some platforms and cameras attach Content Credentials, a standard () that records where a file came from and how it was edited. It helps when present, but it can be stripped, so a missing label does not mean fake.
A worked example
You receive a voicemail in your cousin's voice: "I lost my wallet abroad, please wire $800." The voice is perfect. You hang up and call your cousin's saved number. She is at home and answers right away. The voicemail was a clone. The check cost one minute.
How a voice or face gets cloned
Voice-cloning software learns the pattern of a person's voice from a recording, then produces new speech in that voice from typed text. Image and video fakes work in a similar way: a model learns what a face looks like and generates new frames. The quality depends on the tool and on how much material it has. Public videos and voice notes are the usual source, which is why scams often target people who post a lot.
A second example: a fake that is not malicious
A classmate posts a photo of a shark in a flooded street, made by an image tool and shared as a joke. Many people believe it for a day. No one meant harm, but it spreads as news. Look for signs: a source you cannot , no other photos of the same event, odd details such as warped text or extra fingers. These signs are fading as tools improve, so the source check matters more than the pixel check. Search for the event in news outlets you trust.
Trade-offs
Detection is not a solution. Tools that claim to spot fakes make mistakes in both directions, and a real photo can be wrongly called fake. People also use the doubt as an excuse, saying a real recording is "just AI". So judge claims by evidence around the media: who posted it first, whether others confirm it, and whether it fits what else is known.
What goes wrong
Two failures are common. One is believing a fake because it matches what you expected. Another is sharing before checking, which spreads it further. Both come from speed. If something makes you angry or afraid, that is the moment to wait.
Try it yourself
Agree on a code word with your family. Then pick a viral image from this week and trace it.
- Run a reverse image search with a search engine.
- Find the earliest post you can.
- Note whether a trusted outlet reports it.
Expect to find some images are older photos with a new caption, not AI at all.
Never do this
Do not make or share fake sexual images or videos of real people. It does serious harm, and it is a crime in many places.
A parent gets a call from what sounds like her daughter, crying, saying she has been in an accident and needs money right away. The parent remembers a talk about voice cloning, hangs up, and calls her daughter's real number. The daughter picks up in the middle of a lecture. A code word agreed at breakfast would have settled it in ten seconds.
Check yourself
You get an urgent voice message from a relative asking for money. What do you do?
Not quite. A cloned voice can sound perfect, so verify through a separate channel you already trust.
In the US, who can hold copyright in an image made entirely by an AI?
Not quite. The US Copyright Office requires human authorship, so purely AI output is not protected, though human contributions can be.
A video has no Content Credentials label. What does that tell you?
Not quite. The standard helps when present, but most media do not have it, and it can be removed.
You got 0 of 3 right.
Go deeper
- Article
US Copyright Office · The official reports on authorship and training data.
- Tool
C2PA · The standard for recording where media came from.
- Article
FTC · How voice cloning scams work and how to respond.
Spotted a mistake or something unclear? Report it on GitHub.