← Back to recent questions

I have spent twenty years writing about the limits of certainty in one form or another. My first book, In the Absence of Certainty, is essentially a long argument that "I am sure" is not the same as "I know". So when a question like this lands in front of me, my first instinct is to push back gently on the framing. The world where nobody trusts a photograph is not new. It is the world we used to live in, before photography, when the only things that counted as evidence were a witness, a signature, or an institution that had something to lose if it lied to you.

We had photographs, audio, and video for about a hundred and fifty years. For most of that span, the recording was the proof. That window is closing. We are going back to an older habit. Trust the source, not the format. Most of the change is less dramatic than the headlines suggest.

What changes day to day

Three things, really. First, family scams. The "Hi Mum, I'm in trouble, I need money now" call in a voice that sounds exactly like your son or grandson is the most common harm AI is doing to ordinary households in 2026. The fix is boring and it works: agree on a callback habit and a family code word. If a panicked call comes in, you hang up and ring the person back on the number you already know, or you ask the code word. Both stop a cloned voice cold. The scams and deepfakes page covers this in more detail.

Second, news and social media. The internal voice that used to say "well, there is video of it" has had to be retired. A clip on a feed is now a claim, not a fact. The practical move is to check whether a real outlet you already trust is reporting the same thing, and to look at the source rather than the share. Reuters, AP, the ABC, and the major newspapers are all running provenance markers on their published images now under the C2PA standard. C2PA is a small piece of metadata that travels with a photo or video and records who shot it, when, and what edits were made. It is starting to appear as a small "verified source" badge in browsers and on news sites. It is not perfect (anything can be re-screenshotted to strip it) but it is the first real piece of infrastructure for trustable media on the open internet.

Third, courts and serious decisions. Police, banks, and courts are quietly rebuilding around provenance chains rather than file authenticity. CCTV is trusted because it came from a sealed camera with a clear chain of custody. A clip pulled from a social feed is treated with much more scepticism than it was five years ago. Insurance claim handling has tightened for the same reason.

What does not change

A lot more than you might expect. In-person conversation is exactly as trustable as it always was. Anything physically signed, witnessed, or notarised still holds. Verified archives like Trove, the State Library, or peer-reviewed journals still work because the trust comes from the institution, not the file format. The everyday currency of professional life (a meeting on a calendar, a contract that goes through DocuSign, a payment from a known account) has not changed.

The other thing that does not change is your own life. Your friendships, your family, your local community, the things you actually do on a given Saturday. None of that runs on AI-generated video. The crisis of trust is a crisis of mediated content, not of lived experience.

What I'd avoid

The biggest trap is the equal-and-opposite reaction: doubting everything. There is a thing called the liar's dividend, where someone caught on a real recording can now plausibly claim the recording is faked. If our default becomes "anything could be fake", that does the bad actors' work for them. The better default is "what is the source, and do I trust them". That is a harder habit, but it is the right one.

The other trap is buying gadgets or apps that promise to detect AI-generated content. Detection runs behind generation, always. The detectors that exist in 2026 are unreliable on anything recent. Spend the same effort on the callback habit, the code word, and the source-checking instead. Cheaper, and it actually works.

The longer view

I keep coming back to the database I have been building for the last few years. Five hundred and sixty-five catalogued IT project failures, hand-checked, sourced where I can find a primary source. Almost every catastrophic failure in there has the same shape underneath. Somebody trusted the wrong source. A consulting report nobody had read. A sign-off chain where the signer had not seen what they were signing for. A status report that flowed up the chain getting cleaner at every level. The technology is incidental. The real failure is always epistemic. Who told you, what did you check, what did you take on trust.

The world that learns to live without trusting the format is, for most ordinary purposes, a world that has had to relearn an older skill the rest of us never properly lost. Trust the source. Be specific about who told you. Notice the difference between "I saw it on a feed" and "a friend who was there told me". The good news is that this is a recoverable skill. The bad news is that it requires effort that most of us had been able to skip for a couple of generations because the recording was doing the work. The recording is no longer doing the work. We have to.

As of May 2026. Tool features and pricing change quickly; if you are reading this much later, check the current state before relying on the specifics.