Week 14: Disinformation and Deepfakes

Welcome to Week 14 of the AI Literacy Course.

Imagine receiving a voice message from someone who sounds like a relative:

“I need money urgently. Please don’t call anyone. Just transfer it now.”

The voice may be genuine, cloned, edited, or taken from another recording. You cannot decide safely from the voice alone.

This lesson teaches you how to investigate suspicious media without spreading it or making accusations that the evidence does not support.

The central rule is:

Pause, investigate, and respond without increasing the harm.

Learning goals

By the end of this lesson, you should be able to:

• distinguish misinformation, disinformation, and malinformation;

• distinguish deepfakes from other misleading media;

• separate media authenticity from claim accuracy;

• apply SIFT to suspicious images, audio, and video;

• combine technical clues with stronger contextual evidence;

• state a cautious conclusion;

• correct or report harmful media without amplifying it.

1. Information can mislead in different ways

Not all false or harmful information has the same origin.

Misinformation

Misinformation is false or misleading information shared without clear evidence that the person intended to deceive.

Example:

Someone shares an old photograph because they mistakenly believe it shows a current event.

Disinformation

Disinformation is false or misleading information created or spread deliberately to deceive.

Example:

Someone changes the date and location of a video and knowingly presents it as evidence of a different event.

Malinformation

Malinformation is genuine information used in a harmful or misleading way.

Example:

A private recording is published to harass someone, or a genuine quotation is removed from its context to change its meaning.

Intent can be difficult to prove. Before calling something disinformation, describe what the evidence actually shows.

It may be more accurate to say:

“This video is from 2022, but the post presents it as a current event.”

That is stronger than making an unsupported claim about the poster’s intentions.

2. What is a deepfake?

Synthetic media is the broad term for images, audio, or video generated or significantly altered using technology.

A deepfake is convincing synthetic or manipulated media that imitates a real person’s identity or makes that person appear to say or do something.

Deepfakes can imitate or alter:

• faces;

• voices;

• expressions and movements;

• speech and lip movements;

• actions or events.

Not every AI-generated image is a deepfake. An imaginary landscape generated with AI is synthetic media, but it is not normally a deepfake because it does not impersonate a real person.

Not every deception uses AI either. Misleading media may involve:

• an old photograph presented as new;

• a real video with a false caption;

• a cropped image that hides important context;

• edited audio;

• a copied social media account;

• a staged event;

• a screenshot with altered text.

Simple manipulation can be as harmful as advanced AI.

3. Ask two separate questions

People often ask:

“Is this real?”

That question is too broad. Ask two questions instead.

Question 1: Is the media authentic?

Has the image, audio, or video been generated, edited, or manipulated?

Question 2: Is the attached claim accurate?

Does the caption correctly identify the person, place, date, event, and context?

These questions are related, but they are not the same.

A genuine video may be shared with a false date or location. The file is authentic, but the claim is misleading.

A clearly labelled AI-generated illustration may communicate a true idea without pretending to document a real event. The image is synthetic, but its use is not necessarily deceptive.

You must investigate both the media and the claim around it.

4. Use PAUSE before you investigate

Suspicious content often tries to produce a fast reaction. It may create fear, anger, disgust, admiration, or urgency.

Strong emotion does not prove that something is false. It is a signal to slow down.

Use PAUSE:

• Pause before reacting or sharing.

• Articulate the exact claim.

• Use SIFT to investigate it.

• Seek independent confirmation.

• Evaluate the evidence and possible harm.

PAUSE does not replace the SIFT method from Lesson 12. It reminds you to use SIFT before suspicious media influences your actions.

5. State the exact claim

Do not begin with a vague question such as:

“Is this video fake?”

Write a precise, checkable claim:

“This video shows person X saying sentence Y at event Z on date D.”

Then ask:

• Who posted this claim?

• What would confirm it?

• What would show that it is wrong or incomplete?

• Which sources are in a position to know?

• What harm could result if people believe or share it?

A precise claim produces a more focused investigation.

6. Apply SIFT to suspicious media

Stop

Pause before liking, commenting, downloading, paying, or sharing.

Notice whether the content is pressuring you to act immediately. Be especially cautious when it involves money, passwords, confidential information, threats, or instructions not to contact anyone else.

Investigate the source

Check who published the media.

Ask:

• Is this the original account or a copied account?

• What does the account normally publish?

• When was the account created?

• Does its username closely imitate another person or organisation?

• Does it link to an official website?

• Is the same claim available through a current, independently located official channel?

Follower counts, viral reach, screenshots of verification symbols, and familiar logos are not proof.

Find better coverage

Look for independent reporting or confirmation from sources that could know what happened.

Do not rely on ten accounts repeating the same unverified post. Repetition is not the same as independent evidence.

Search for:

• the names of the people or organisations involved;

• distinctive words from the quotation;

• the stated event, location, and date;

• reporting from credible local or specialist sources;

• statements published through independently located official channels.

Trace the media to its original context

Try to find the earliest available version.

Useful methods include:

• reverse image search;

• searching keyframes from a video;

• quotation search;

• archive search;

• comparing different versions of the same file;

• checking the date, location, weather, language, or visible landmarks.

An earlier version may show that an image comes from another year, country, or event.

A failed search proves very little. Search tools may not find new, private, deleted, cropped, or heavily edited material.

7. Combine different kinds of evidence

No single clue should decide the investigation.

Source and context

Consider:

• who first published the media;

• where and when it appeared;

• how it reached you;

• whether the caption matches the original context;

• whether independent sources confirm the event.

Technical clues

Possible clues include:

• speech and lip movements that do not match;

• sudden changes in voice quality;

• unnatural timing;

• inconsistent shadows or reflections;

• warped hands, teeth, jewellery, or background details;

• visible editing seams;

• unusual file information.

These are reasons to investigate further—not proof of manipulation.

Compression, poor lighting, video filters, translation, frame loss, and ordinary editing can make authentic media look strange. High-quality manipulation may also contain no obvious visual defects.

Detection tools

Some tools estimate whether media may have been generated or altered by AI. Their results are uncertain.

A detector score is not a verdict because:

• different tools may give different results;

• real media can be incorrectly flagged;

• manipulated media can escape detection;

• performance changes as generation technology develops.

Use detector results only as one clue alongside source tracing, context, and independent confirmation.

Provenance

Provenance means information about where a file came from and what happened to it.

Metadata, content credentials, camera records, editing history, or newsroom records may strengthen confidence when they are available and valid.

However:

• metadata can disappear when a file is copied or uploaded;

• screenshots may remove provenance information;

• information can sometimes be altered;

• missing credentials do not prove that a file is fake.

Provenance can support a conclusion, but it cannot replace a wider investigation.

8. Verify urgent requests independently

A cloned or edited voice may sound familiar. A video call may also appear to show someone you know.

Do not rely on familiarity alone when a request involves money, personal data, passwords, or confidential information.

Instead:

1. End or pause the conversation.

2. Use a phone number or contact route you already know.

3. Contact the person or organisation independently.

4. Use a second verification method when possible.

5. Do not transfer money or reveal sensitive information until the request is confirmed.

Do not use contact details supplied inside the suspicious message.

If a payment may already have been made, contact the bank or payment service quickly through its official channel.

9. Avoid the liar’s dividend

The existence of deepfakes creates another danger: people can dismiss authentic evidence by simply claiming that it was generated by AI.

This is called the liar’s dividend.

Responsible verification must protect against two mistakes:

• accepting manipulated media as authentic;

• rejecting authentic media as fake without sufficient evidence.

Do not treat suspicion as proof. An accusation of fakery can also damage innocent people.

10. State only what the evidence permits

Your conclusion should reflect the quality of the evidence.

Possible conclusions include:

• Supported: several strong and independent sources confirm the claim.

• Misleading context: the media is genuine, but the caption, date, location, or interpretation is inaccurate.

• Manipulated: reliable evidence shows that the media was significantly altered.

• Synthetic: reliable evidence shows that the media was generated.

• Unverified: available evidence does not currently confirm the claim.

• Insufficient evidence: there is not enough reliable information to decide.

Explain what you know, what you do not know, and how confident you are.

“Not yet verified” is often more accurate than “definitely real” or “definitely fake.”

11. Respond without increasing the harm

Do not repost harmful media merely to criticise it. Reposting can expose more people to the original claim and may further harm the people shown.

A safer response may include:

• not sharing the material;

• preserving the link or file privately when evidence is needed;

• warning an affected person through a trusted channel;

• reporting an impersonating account or deceptive post;

• providing a correction with reliable context;

• linking to independent evidence instead of repeating the harmful media;

• avoiding unnecessary names, images, or personal details.

The appropriate response depends on the risk.

Fraud, threats, intimate imagery, harassment, or immediate danger may require help from a trusted adult, school, employer, platform, bank, support organisation, or relevant authority.

Do not investigate in a way that puts you or another person at risk.

Classroom activity: Investigate without amplifying

Use a harmless example selected by your teacher. Do not repost the material publicly.

Complete a verification record containing:

1. Exact claim: What does the post claim happened?

2. Initial reaction: What emotion or urgency might influence the audience?

3. Source: Who published it, and is this the original source?

4. Context: What date, place, people, and event are claimed?

5. Three checks: Which independent methods did you use?

6. Evidence found: What supports or weakens the claim?

7. Missing evidence: What remains unknown?

8. Conclusion: What does the evidence currently permit you to say?

9. Confidence: How certain are you, and why?

10. Safest response: Should you ignore, correct, report, or seek help?

Present your findings as a short written analysis, audio response, or brief presentation.

Reflection questions

1. Why is “Is this real?” often too broad?

2. Can authentic media still support a false claim? Give an example.

3. Why should technical defects be treated as clues rather than proof?

4. What evidence is stronger than a detector score?

5. Why is intent difficult to establish?

6. How can you verify an urgent voice request?

7. What is the liar’s dividend?

8. When is “not yet verified” the most responsible conclusion?

9. How can a correction accidentally amplify harmful content?

10. What would be the least harmful response in your example?

Key vocabulary

Misinformation:
False or misleading information shared without clear evidence of an intention to deceive.

Disinformation:
False or misleading information deliberately created or spread to deceive.

Malinformation:
Genuine information used in a harmful or misleading way.

Synthetic media:
Images, audio, or video generated or significantly altered using technology.

Deepfake:
Convincing synthetic or manipulated media that imitates a real person or makes that person appear to say or do something.

Provenance:
Information about where media came from and what happened to it.

Liar’s dividend:
The opportunity to dismiss authentic evidence as fake because synthetic media exists.

Amplification:
Increasing the reach or visibility of information, including information being criticised.

Summary

In Week 14, we learned that suspicious media cannot be judged by appearance alone.

First, pause and state the exact claim. Then apply SIFT: investigate the source, find independent coverage, and trace the media to its original context.

Separate two questions:

• Is the media authentic?

• Is the attached claim accurate?

Combine source history, context, independent confirmation, technical clues, and provenance. Treat detector results as uncertain clues, not final answers.

Finally, state only what the evidence supports and respond without spreading the harm further.

Pause, investigate, and respond without increasing the harm.

Lesson 14 Interactive Quiz: Disinformation and Deepfakes

Choose one answer for each question. Then select Check my answers. This practice quiz does not collect names or scores.

1. A genuine photograph from 2021 is shared as evidence of an event happening today. What is the main problem?

A. The photograph must be AI-generated B. Authentic media is being presented with misleading context C. Old photographs cannot be shared online D. The photographer has proved deliberate deception

2. You confirm that a video has a false caption, but you cannot determine why the person shared it. What is the most responsible conclusion?

A. The person definitely created disinformation B. The video itself must be synthetic C. The false caption does not matter because the video is genuine D. The caption is false, but the person’s intention has not been established

3. Which example is best described as a deepfake?

A. Synthetic video that makes a real public figure appear to deliver a speech they never gave B. An AI-generated picture of an imaginary city, clearly labelled as fictional C. A photograph made brighter with ordinary editing software D. A genuine interview shortened for a news summary without changing its meaning

4. Which is the most precise and checkable claim?

A. This video looks suspicious B. Something about this speaker seems false C. This video shows Minister X making statement Y at conference Z on 12 May D. People online say this event never happened

5. Which investigation provides the strongest evidence about a suspicious video?

A. Deciding that the speaker’s blinking looks unnatural B. Finding the original recording and checking it against independent reporting C. Trusting the post because it has thousands of shares D. Using one AI detector and accepting its score as final

6. An AI detector says an audio recording is “78% likely to be generated.” What should you do?

A. Announce that the recording is definitely fake B. Accept it as authentic because the score is below 100% C. Test it repeatedly until one tool gives the answer you expect D. Treat the score as one clue and continue investigating other evidence

7. A familiar voice calls and urgently asks you to transfer money. What is the safest first response?

A. Pause and contact the person through a number you already know B. Transfer a small amount first to test the request C. Trust the request because the voice sounds familiar D. Use the new contact details provided by the caller

8. A person rejects a verified recording by saying, “Anything can be made with AI now.” Which concept does this illustrate?

A. Provenance B. Malinformation C. The liar’s dividend D. Reverse image search

9. You cannot find the original file, and reliable sources have not confirmed or rejected the claim. Which conclusion is justified?

A. Definitely authentic B. Not yet verified C. Definitely manipulated D. Probably true because no one has disproved it

10. You confirm that a harmful viral clip has a false caption. What is the most responsible response?

A. Repost the complete clip so everyone can inspect it B. Name every person shown so they can defend themselves C. Add an angry comment to attract attention to the correction D. Provide reliable context without redistributing more of the harmful material than necessary

Reflection: When suspicious media appears, ask two questions: “Is the media authentic?” and “Is the claim attached to it accurate?”