Talking to Teens About AI and Deepfakes: Conversation Starters for Parents and Educators

If a teenager in your life receives a video call from someone claiming to be a government officer, hang up. Call the agency back on a number you look up yourself. That is the first instruction. It matters more than any lecture about AI.

Asking young people to spot synthetic media by eye is the wrong move. The technology is already past the point where visual inspection works. A 2024 Institute of Policy Studies survey found that 62% of Singapore residents are concerned about their ability to distinguish manipulated media from genuine recordings, and they are right to be. The goal is not to train a human detector. The goal is to teach a verification habit that works regardless of what the next generative model produces.

That habit is lateral reading. It does not require knowing how Midjourney, DALL-E, or Sora work internally. It requires leaving the material you are unsure about and opening new tabs to check the source, the claim, and the context. That skill survives every model update. The conversation starters below give a teenager something they can do, not something they should believe.

The most effective opener is not a lecture. It is a specific thing to look at together. Pick an example recent enough to feel relevant but not so current it is still being debated. The Singapore Police Force reported at least five scam cases in the first half of 2024 involving digitally altered video calls that impersonated officials. That is a concrete case. Ask: what would you check if your phone rang with a video call from a number you did not recognise, and the person on screen looked like a government officer?

Start With The Audio, Not The Video

AI voice cloning is further along than most people realise. The Singapore Police Force identified a specific voice scam method in 2024: AI-generated voice clones of family members requesting urgent money transfers. Ask a teenager what they would do if they received a panicked call from a relative asking for cash. The answer is not to listen harder. Hang up and call the person back on a number they already have saved. That is a verification reflex, not a listening skill.

Show Them A Cheapfake First

A cheapfake is easier to understand than a synthetic AI creation. It is media altered with simple tools: a video slowed down, a photo cropped, an old image labelled as new. The fabrication method is different but the intent is the same. Start with a cheapfake so the teenager sees that the problem is not limited to AI. Once they grasp that any media can mislead, the conversation about manipulated media has a foundation.

The core concept to get across is that looking real is no longer evidence of being real. A 2024 Ministry of Education survey found that 54% of secondary school students were confident they could identify AI-generated material. Confidence and accuracy are different things. The NLB S.U.R.E. programme's updated framework, S.U.R.E. for AI, released in July 2024, provides four steps that work without technical knowledge: Source (check the origin), Understand (know how AI works), Research (verify with other sources), Evaluate (assess fairness and intent). That is the framework. It is not a list of visual tells that will be obsolete in six months.

What To Say About Tools Like Midjourney And Sora

You do not need to explain how diffusion models work. You need to say that these tools can now produce photorealistic images and video of things that never happened. The Singapore Police Force reported that a single scam victim lost an amount ranging from modest to life-changing in 2024, with the exact figure depending on the scam type. Check the SPF website for the latest breakdowns. The financial stakes are real. The conversation is about consequences, not technology.

The C2PA Content Credentials Standard And Its Limits

C2PA credentials are an emerging standard that attaches a digital provenance record to media. If an image or video has a C2PA credential, you can check when it was created, by which device or tool, and whether it has been edited. The honest thing to tell a teenager: adoption is low. Most material you encounter will not carry a credential. C2PA is a future tool, not a present solution. The verification habit matters more.

The skills that last are not visual tells. They are behaviours. The National Library Board's S.U.R.E. programme recommends reverse image search via Google Images and TinEye as a detection technique. It works because it checks where an image has appeared before, not what the image looks like. If a supposedly new photo of a local event turns out to have been posted on a foreign site three years ago, the teenager has an answer without needing to judge image quality.

How Lateral Reading Replaces Spotting

Lateral reading is the single most durable verification technique. Instead of studying the material itself, the reader opens new tabs to check the source. If a forwarded message on WhatsApp claims to be from a Singapore government agency, the teenager opens a tab to the actual agency website and looks for the advisory there. If it is not there, the message is unverified. That takes thirty seconds and requires no AI knowledge.

The Forwarded Message Trap

WhatsApp labels messages that have been forwarded multiple times. That label is a structural feature that will persist. Teach a teenager that a forwarded-as-received message has no original source attached. The first step is to ask who wrote it. If the answer is unknown, the claim cannot be trusted. That is not scepticism. That is basic source checking.

Teenagers need to know that Singapore has specific laws about AI-generated material. The Protection from Harassment Act (POHA) covers the distribution of manipulated images that cause harassment, alarm, or distress. Section 3 carries a penalty of up to 12 months imprisonment or a fine of up to S$5,000 or both. That includes non-consensual synthetic images. In February 2025, the Criminal Procedure (Miscellaneous Amendments) Act 2025 criminalised the distribution of non-consensual intimate synthetic images with a penalty of up to five years imprisonment. A teenager who understands the law treats this as a serious offence is a teenager who will think twice before sharing or creating such material.

What To Say About Non-Consensual Synthetic Images

Do not avoid this topic. The Singapore Police Force reported 55,810 scam cases in 2024, and manipulated media is increasingly used in impersonation and harassment. The conversation should be direct: creating or sharing a non-consensual synthetic image of anyone, including a classmate, is a criminal offence under POHA and the Penal Code. Legal recourse exists. The teenager should know that the Community Justice Centre provides legal aid for victims, and the Ministry of Home Affairs' Victim Care Cadre Programme, launched in 2024, includes support for online harm victims.

The Digital Footprint Reality

A digital footprint is permanent. Material posted online can be archived, screenshotted, or indexed even after deletion. That includes AI-generated images and video. The Ministry of Education's updated cyber wellness curriculum, refreshed in 2024, includes AI literacy components taught in Character and Citizenship Education lessons. The teenager should understand that sharing a synthetic image is not a victimless act, and the legal consequences do not disappear because the material was created with a tool rather than a camera.

This conversation method suits parents and educators who need concrete starters, not a lecture script. It suits the everyday Singaporean who receives forwarded messages and wants a repeatable checklist. It suits the curious young person who already knows some internet culture but wants depth without condescension. It does not suit a reader looking for political commentary or activist calls to action. It does not suit someone who wants official government communications or legal advice. It does not suit a reader outside Singapore seeking general media literacy theory without local laws and tools. The most common failure is the AI realism trap: believing that because something looks real, it must be real. That belief is what these conversation starters are designed to break.