The Future with Hannah Fry episode 3

The Future with Hannah Fry episode 3

The Future with Hannah Fry episode 3 confronts one of the most urgent questions of the digital age: does the vast ocean of data we generate every day make society safer, or does the death of privacy leave us dangerously exposed? Mathematician and writer Professor Hannah Fry explores both sides of that question with unflinching clarity, moving from the unsettling mechanics of online harassment to the life-or-death stakes of open-source war crimes investigation. The answer, as Fry demonstrates, is neither simple nor comfortable — and it affects every person who has ever sent a message, opened an app, or posted a photo online.


What makes this episode so striking is the scale of what has already been given away. For over 20 years, people have been pouring personal information into the digital world — not just through deliberate sharing, but through every tiny interaction: a tap on a phone screen, a cookie agreement accepted in irritation, a mailing list sign-up at a conference. Each of these generates what Fry calls “little breadcrumbs,” data signatures that reveal where you are and what you are doing. Individually, they seem harmless. Collectively, they constitute something closer to a permanent record of your life.

That record can be read by anyone who knows how to look. And as The Future with Hannah Fry episode 3 makes clear, the techniques required are no longer the preserve of intelligence agencies or elite hackers. They are free, accessible, and increasingly automated.



One of the episode’s most arresting demonstrations involves geolocation — the process of pinpointing exactly where and when a photograph or video was taken, using nothing but the image itself and an internet connection. Fry walks through the methodology with the precision you would expect from someone who studies patterns in human behaviour for a living.

The process starts with what is visible in the background. Distinctive buildings can be identified through a reverse image search, trawling the entire internet for prior uploads of the same structure. Beyond architecture, the clues multiply quickly. Plug sockets identify the country. Road signs and signposts narrow the region. Even the species of trees in frame can help eliminate possibilities. Once a location is confirmed, the length of shadows in the image can determine the approximate time of day the photograph was taken. Free online tools perform many of these calculations automatically.

The critical point Fry emphasises is accumulation. A single clue may be circumstantial. But layer enough of them together — building, socket, sign, tree, shadow — and the result becomes, in her words, irrefutable. You know exactly where and when a photograph was taken. You cannot fake the shadow length. You cannot retrofit the plug socket. The evidence is written into the image whether the subject intended it or not.

This matters enormously, and not just as a party trick. Geolocation is one of the foundational techniques used by investigative journalists to verify footage from conflict zones, expose fabricated propaganda videos, and document war crimes. It is also a core component of the accountability work that organisations like Bellingcat conduct daily — placing an event precisely on a map, at a precise moment in time, using evidence that cannot be credibly denied. It is also, however, a tool that can be turned against private individuals with devastating effect.

The Future with Hannah Fry episode 3

Doxxing, Swatting, and the Playbook of Online Harassment

The last decade has produced a notable rise in what is known as doxxing: the practice of using internet searches to uncover personal information about someone, then sharing it publicly to encourage others to cause harm. Fry traces how these attacks typically begin with an online disagreement before escalating rapidly. Aggressors join the social media groups their targets belong to, infiltrate online communities connected to them, spread false accusations, and work to isolate the victim from their own networks.

The attacks do not stay digital. One particularly dangerous technique, known as swatting, involves calling the authorities with fabricated threats of violence at the victim’s home address, prompting armed police to storm the property. Some of these incidents have ended in the victims’ deaths — with those deaths then celebrated in online forums by the very people who orchestrated the attacks.

The Future with Hannah Fry episode 3

Fry is precise about the current scale: these extreme cases remain, mercifully, relatively rare. The labour involved in a sustained doxxing campaign limits how many can be executed simultaneously. The key word, however, is “currently.” Because the third major thread running through The Future with Hannah Fry episode 3 concerns what happens when these techniques no longer require human effort at all.

Large Language Models and the Industrialisation of Manipulation

Fry meets Rishi Bommasani, a researcher who studies the benefits and dangers of large language models — the algorithms underpinning a new generation of AI chatbots capable of generating realistic, conversational text from a simple prompt. The demonstration is deliberately mundane and therefore all the more unsettling. Given a few basic personal details — a name, a hobby, a social habit — the AI produces a tailored persuasive message that slots those details naturally into its argument, making it feel personal rather than machine-generated.

Bommasani describes the training data behind these models as many thousands of human lifetimes’ worth of text. The entire readable internet, essentially, compressed into an algorithm that can produce passably human prose on demand. For legitimate uses — sophisticated customer service bots, plain-English internet search, writing assistance, translation of language into computer code — this is genuinely transformative technology. Bommasani acknowledges there is a great deal that is exciting and positive about how these models can augment human capabilities.

The danger emerges precisely from the same capability. Large language models cannot distinguish between truth and fiction. They generate confident, fluent text regardless of whether the underlying claims are real. That limitation matters enormously when the intention is not accuracy but persuasion. Bommasani is direct: these tools unlock something that disinformation actors have never had before — the ability to generate misleading text at industrial scale. Hiring a human to write targeted propaganda is expensive and slow. Asking a machine to produce thousands of personalised variations, each calibrated to a specific individual’s known preferences and vulnerabilities, is neither.

Fry presses the point further. The digital footprint each person leaves online is vast. Posts, comments, purchases, event registrations, group memberships — all of it paints a portrait of beliefs, anxieties, and susceptibilities. A large language model capable of reading that portrait could, in theory, generate a message specifically engineered to exploit whatever makes a particular person most likely to be misled. Not just fluent text. Targeted text. Personalised deception, automated and scalable.

Bommasani frames it in terms of existing criminal behaviour. Spam, fraud, and financial scams already use publicly available personal data to make approaches feel credible. A language model doesn’t change the underlying motivation — it simply removes the human bottleneck. The same attack that once required a skilled operator working case by case can now be replicated across millions of targets simultaneously, each version individually calibrated, each one sounding like it was written specifically for the person receiving it. That is not a future risk. The components are already in place.

The Fragmentation of Reality as a Political Weapon

The logical endpoint of that capability is what makes The Future with Hannah Fry episode 3 genuinely alarming. If you can build a targeted information bubble around one individual automatically, you can do it around every individual in a country simultaneously. Each person receives a different version of events, shaped to reinforce their existing fears and convictions. Nobody is sharing the same reality. Nobody can agree on what is true.

Fry articulates the consequence plainly: you can create fragmented realities that split a country apart without ever having to fire a weapon. That is not a science fiction scenario. The technical components already exist. The data is already out there, accumulated over decades and growing more potent, not less, as the algorithms trained on it become more sophisticated. There is no reversing what has already been published. There is no reclaiming data already surrendered.

The political implications extend beyond elections and referendums. Trust in shared institutions — courts, public health bodies, journalism, democratic processes — depends on a minimum level of shared factual ground. When that ground fractures, and when the fracturing can be engineered deliberately and cheaply, the consequences reach far beyond any individual targeted by harassment. An entire society’s capacity for collective decision-making is at stake.

Bommasani describes this as one of the central unsolved problems of the field: how should society benefit from these technologies in a broader sense, and how do we ensure they are being used ethically? These are not questions with easy answers. But they are questions that demand urgent attention, because the gap between asking them and the moment when they become practically urgent is narrowing fast.

Open-Source Intelligence and the Case for Ukraine

At the exact midpoint of the episode, The Future with Hannah Fry episode 3 pivots to a counter-argument — and it is a powerful one. The same tools that enable harassment and propaganda are also enabling accountability at a scale and speed previously impossible.

Fry meets Eliot Higgins, founder of Bellingcat, an investigative journalism organisation that describes itself as an intelligence agency for the people. Higgins began as a hobbyist, just over a decade ago, before building what is now a team of more than 30 staff. Their method is digital forensics conducted entirely with publicly available information: photographs, YouTube videos, social media posts, satellite imagery. No secret sources. No classified access. Just open-source data and the skills to read it.

In Ukraine, Bellingcat’s work has been immediate and consequential. From the first day of the conflict, they began building a database of geolocated incidents — each circle on their map representing a location where civilians were injured or killed. The purpose is documentation: creating a verifiable, cross-referenced record of events that can be used by NGOs, human rights organisations, and international accountability bodies.

The internet, Higgins explains, has become a receiver for information from incidents happening on the ground. Smartphone cameras, social media distribution, and satellite imagery together make it very hard to hide things that once would have disappeared without record. Even events not directly captured on video or photography leave traces — the satellite record of the physical world continues regardless of whether anyone on the ground is filming.

Satellite Evidence, the Mariupol Theatre, and What Cannot Be Denied

The Mariupol theatre bombing stands as one of the most chilling examples of what open-source evidence can establish. Satellite imagery taken before the strike clearly showed the word “Children” written in large letters on the ground outside the building — a standard warning to aircraft that civilian non-combatants, specifically children, were sheltering inside. The building was bombed regardless. That satellite image constitutes evidence of a specific, deliberate choice. Accountability, Higgins says, can come in many different forms, but it begins with documentation that cannot be credibly disputed.

This is the essential distinction that Bellingcat embodies: the shift from assertion to verification. Their work does not rely on anonymous tips or unverifiable testimony. Every conclusion is traceable back through a chain of openly available evidence that anyone can examine. That transparency is, in Higgins’s view, both their methodology and their message. It is not about secret sources. It is about sharing the information, sharing the reasoning, and inviting scrutiny.

The MH17 Investigation and the Missile Launcher Trail

One of Bellingcat’s most significant investigations demonstrates exactly how meticulous this methodology can be. Following the downing of Malaysia Airlines Flight 17, they used photographs and videos posted to social media to trace the journey of the suspected missile launcher from Russian territory, across separatist-controlled borders, to the location from which the aircraft was destroyed. They cross-referenced dents in the vehicle’s side skirts across multiple photographs to confirm they were tracking the same launcher throughout. The physical damage on the machine’s bodywork functioned as a fingerprint, making the case for Russian involvement documented rather than merely alleged.

The breadth of Bellingcat’s investigations extends well beyond Ukraine and Russia. They have investigated US air strikes in Syria, including one targeting a building claimed by authorities to be an Al-Qaeda meeting location but shown by Bellingcat’s analysis to be a mosque used by civilians. They have used the unique stripe patterns on tiger cubs’ foreheads — functioning, in effect, as fingerprints — to match animals appearing in Instagram photo shoots back to someone illegally supplying them to social media influencers. The same logic, applied to vastly different problems. Open data, methodical comparison, irrefutable visual evidence.

Fabricated Propaganda and the Metadata That Exposes It

Higgins and his team are equally active in debunking fabricated content circulating around the Ukraine conflict. Russia published footage claiming to show Ukrainian operatives attacking chlorine storage tanks to poison civilians in separatist territories. Bellingcat examined the video’s metadata — the hidden information embedded in the file itself about how and when it was created.

The project folder’s creation date was a full year before the footage was purportedly filmed. Within the file, a reference to a YouTube video with an almost identical title surfaced. When they checked, the audio of the supposed chlorine explosion matched, down to the waveform, a publicly available clip on YouTube. The explosion sound had been imported directly from an existing video and edited over the new footage.

Higgins is clear about the intent: it was meant to feed a narrative that Ukrainians would be willing to use chemical weapons against civilians. The video was a fabrication. And Bellingcat proved it using nothing more sophisticated than the metadata the video’s creators left in the file and a YouTube search.

The Flattening of Power and Why Citizens Now Have Agency

What strikes Fry most about Bellingcat’s work is not any single investigation but what the organisation’s existence signifies about the redistribution of power in the digital age. Ten or fifteen years ago, these investigations simply could not have been conducted this way. The data did not exist in publicly accessible form. The tools for analysing it were not available to civilians. The networks for coordinating amateur researchers around a single problem had not yet formed.

Now they have. Higgins describes it as a flattening of power structures. Governments, militaries, and established media organisations once held near-exclusive control over what information reached the public. A credentialled reporter with secret sources could investigate; everyone else could not. Now anyone can start a website, build a following, and contribute meaningfully to an investigation simply by geolocating a single video. In Ukraine, ordinary Twitter users have done exactly that, each adding one small piece to a much larger picture.

Fry finds genuine cause for optimism here. The very same transparency that makes individuals vulnerable to harassment and surveillance also makes it harder for governments and armed forces to commit atrocities unobserved. There is nowhere on the globe, she notes, where there is not a camera pointing, an audio recording being made, or a satellite image that can be traced. That exposure cuts both ways. It cuts against the private individual whose location can be pinpointed from a single photograph. But it also cuts against the general who orders a strike on a civilian shelter and believes the evidence will be buried.

Privacy Lost, Power Gained, and the Shape of What Comes Next

The Future with Hannah Fry episode 3 does not resolve the tension it identifies. It cannot, because the tension is real and ongoing. We have structured all of our communication in ways that leave us open to exploitation, Fry says directly. There is no undoing that. The data already exists. The algorithms trained on it grow more capable every year. The techniques for geolocation, personalised manipulation, and mass-scale disinformation production are not going to become less powerful or less accessible.

And yet. The same forces that create that exposure have also created Bellingcat, and the volunteers who geolocate atrocity footage from their home computers, and the researchers working to understand how large language models can be made safer. Fry closes with a measured but real sense of optimism — not because the threats are smaller than they appear, but because ordinary people now hold tools that were unavailable to previous generations. We are not anonymous any more.

That is disconcerting, and it is also true. But in exchange, we have gained something: the ability to hold power to account in ways that were simply impossible before. Atrocities that would once have been denied, buried, or simply never recorded now leave traces that patient, methodical citizens can find and verify. Propaganda videos that would once have circulated unchallenged can now be dismantled in hours using nothing but the file’s own metadata and a YouTube search.

Whether that exchange is a net gain depends on what we build next.

FAQ The Future with Hannah Fry episode 3

Q: What is geolocation and how can it reveal where a photo was taken?

A: Geolocation is the process of pinpointing exactly where and when a photograph or video was taken using only visual clues in the image. Investigators examine background buildings, plug socket types, road signs, tree species, and shadow lengths. Each clue narrows the possibilities, and when layered together they produce irrefutable evidence of a precise location and time — using nothing more than an internet connection and free online tools.

Q: What is doxxing and why is it so dangerous?

A: Doxxing involves running internet searches to uncover someone’s personal information, then publishing it publicly to encourage others to cause harm. Attacks typically begin with an online dispute before escalating. Aggressors infiltrate the victim’s social media communities, spread false accusations, and work to isolate them. In extreme cases, attackers call authorities with fabricated threats — a technique known as swatting — which has led armed police to storm victims’ homes and, in some cases, resulted in deaths.

Q: How do large language models make disinformation more dangerous?

A: Large language models can generate realistic, persuasive text at industrial scale — something disinformation actors have never previously had access to. Hiring humans to write targeted propaganda is expensive and slow. A language model produces thousands of personalised variations automatically, each calibrated to exploit a specific individual’s known beliefs and vulnerabilities. Because the generated text is fluent and convincing, people can be misled into making wrong decisions without ever suspecting the message was machine-generated.

Q: Can AI really target individuals with personalised manipulation?

A: Yes — and the data required already exists. Every social media post, group membership, purchase, and event registration creates a profile of a person’s beliefs, preferences, and anxieties. A large language model trained on that profile can generate messages engineered specifically to exploit whatever makes that individual most susceptible to being misled. Researcher Rishi Bommasani demonstrated this directly: given a name and a hobby, an AI produced a tailored persuasive argument that felt personal rather than automated.

Q: What does it mean to split a country apart using data without firing a weapon?

A: If a language model can build a targeted information bubble around one individual automatically, it can do the same for every individual in a country simultaneously. Each person receives a different version of events, tailored to reinforce their existing fears. Nobody shares the same factual reality. Trust in shared institutions collapses. This fragmentation of public reality represents a form of political warfare that requires no military force — only data, algorithms, and scale.

Q: What is Bellingcat and how does it investigate war crimes?

A: Bellingcat is an investigative journalism organisation founded by Eliot Higgins that describes itself as an intelligence agency for the people. It conducts digital forensics using entirely open-source material — social media photographs, YouTube videos, satellite imagery, and geolocation data. No classified access or secret sources are required. In Ukraine, Bellingcat built a geolocated database of civilian casualty incidents from day one of the conflict, making verified records available to human rights organisations and international accountability bodies.

Q: How did satellite imagery prove intent in the Mariupol theatre bombing?

A: Satellite images taken before the strike clearly showed the word “Children” written in large letters on the ground outside the Mariupol theatre — a recognised warning to aircraft that civilians, specifically children, were sheltering inside. The building was bombed regardless. That satellite record constitutes documented evidence of a deliberate choice, making it a critical tool for war crimes accountability because it cannot be credibly disputed or retroactively altered.

Q: How did Bellingcat link Russia to the downing of Malaysia Airlines Flight MH17?

A: Bellingcat used photographs and videos posted to social media to trace the suspected missile launcher’s journey from a Russian air defence base, across separatist-controlled borders, to the launch location. Crucially, they cross-referenced dents in the vehicle’s side skirts across multiple images to confirm they were tracking the same launcher throughout. Physical damage on the vehicle acted as a fingerprint, turning publicly available social media posts into a chain of documented evidence.

Q: How did investigators expose a Russian propaganda video about chlorine tanks in Ukraine?

A: Bellingcat examined the video’s embedded metadata and found the project folder had been created a full year before the footage was supposedly filmed. Inside the file, a reference pointed to a YouTube video with an almost identical title. Checking that clip revealed the explosion audio matched frame-for-frame, down to the waveform — the sound had been copied directly from an existing YouTube upload. The video was a fabrication built to support a false narrative about Ukrainian chemical attacks.

Q: Does the loss of online privacy mean ordinary people have lost power, or gained it?

A: Both. As individuals, the inability to remain anonymous is genuinely disconcerting — a single photograph can pinpoint your location, and your data profile can be used to manipulate you. However, the same transparency makes it far harder for governments and armed forces to commit atrocities unobserved. Organisations like Bellingcat, and ordinary citizens geolocating conflict footage from home computers, represent a meaningful redistribution of investigative power away from institutions and toward the public.

Tags: , , , , , ,

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top