Privacy has always mattered and will matter even more in the AI age.

But there are people arguing that we live in a post-privacy world.

They say that people do not care about data privacy as long as the product delivers for them.

And on some level, they are probably right.

With so many digital products, data is collected by default, including in ways that are not immediately obvious. When you visit a website or use an app, a whole trove of data is generated and collected about you; what you do on the website or app, the other apps and websites you use, your location, and other bits of information constituting your digital footprint.

This data are used to target you with ads, sold to data brokers and even used to train the algorithms that decide what you see, how much you pay, and sometimes whether you qualify for a service at all.

This processing happens in the background. It consists of covert operations hidden away from the frontend experience that users are predominantly exposed to. What goes on under the hood has never really been their concern.

But AI changes this user experience significantly.

Think about how ads are deployed on a social media platform vs ads deployed inside an AI chatbot.

On social media, ads are mainly based on the data users provide passively. Platforms make inferences about their interests based on user activity and predict the products and services that are most aligned with those interests.

Inside an AI chatbot, the experience is far more intimate. Heavy users treat these systems as trusted companions: they bounce ideas off them, confide their private thoughts and feelings, and use them to organise their lives.

That intimacy makes advertising feel different. The ads are generated from the user’s own conversations and appear inside the thread itself, woven into an exchange the user experienced as private. They land as alien intrusions, disruptive in themselves, but also revealing. Each one is a reminder that someone was listening all along, and that the conversation was never really private.

Privacy stops being an abstract worry and becomes an immediate one, in a way few other digital products manage to provoke. The ‘off’ switch for personalised ads no longer feels like a preference buried in settings. It feels like a necessary act of reclamation, a way to restore some quiet to a space that was supposed to be yours.

And accordingly, consumers are unlikely to adopt AI products and services that do not provide the necessary privacy-preserving measures.

This will be the same for ambient AI agents that work in the background autonomously.

In this context, AI is being pedestalised as the intelligence engine sitting between all your different stores of data. If you are a consumer, it is your emails, messages, calendar, bank account.

Nobody hands over that kind of access without reassurance. The moment you start connecting data sources, whether onboarding an agent or building a workflow, three questions surface immediately: Will this data be handled reliably? Will confidential information stay confidential? And what recourse do I have when something goes wrong? Products that cannot answer these questions convincingly will not earn users’ trust.

But this requirement will not just be the expectation of individual consumers - enterprises will also have this on their checklist before procuring an AI system.

Before a business uses an AI product or service, they want to ensure that their data are not used for model training, that there is limited or zero data retention, that data residency controls exist and that the confidentiality of communications will be maintained, among other things.

Privacy will no longer be something that can be ignored, treated as a nice aesthetic, or bolted on in a rush. It will become a primary expectation, and AI providers who have not built it into their products will lose out. Consumers and enterprises simply will not use AI that handles data carelessly.

We are not in a post-privacy era. We are heading toward the opposite.

Cursor might be paving a way for the so-called wrapper companies to actually survive and thrive.

Earlier this year in March the maker of its popular AI-powered IDE announced its agentic coding model called Composer 2, designed for long-horizon coding tasks. And in May it announced Composer 2.5, which according to Cursor is better, more reliable and more pleasant than its predecessor.

Prior to the development of its Composer series of models, Cursor relied on coding models developed by third parties, including those from frontier labs like OpenAI and Anthropic. The release of the original Composer in October 2025 model marked the first time that the company offered its own proprietary model inside its app.

The motivation for developing Composer was to make a more efficient model built specifically for coding tasks inside the Cursor app. While models from frontier apps work well for coding purposes, they are general-purpose models built to do many different things. This means that the model weight’s are trained to work across a broad range of tasks rather than being optimised for coding specifically.

Cursor therefore wanted to build its own model where all the weights would be optimised for coding for its specific app and also much cheaper to run.

Wrapper companies building proprietary models was thought to be rather infeasible given the fierce competition from frontier labs and the enormous resources needed to build such things; the compute, the data, the talent and the money to fund it all.

But Cursor is demonstrating that this may not always be true. The company has been building its AI-powered IDE for a number of years, over which time it has accumulated money, talent and, crucially, data - code, conversation threads and other user activity that Cursor can rely on to build its own model.

This is described in its privacy policy. Obviously if you use Cursor then the app will be processing what the policy defines as “Inputs and Suggestions”, which means the prompts and code submitted to Cursor and the AI-generated responses to these inputs. Cursor only uses this data to train its models if users explicitly agree to Cursor using this data for such a purpose (and this is also reiterated in its terms of service). Cursor’s Data Use & Privacy Overview page describes how this is determined by the Privacy Mode settings in the Cursor app - if this is switched off, then Cursor “may use and store codebase data, prompts, editor actions, code snippets, and other code data and actions to improve [its] AI features and train our models.”

It is this kind of data that Cursor can leverage to build a model like Composer. To build a specialised model built for long-horizon coding tasks inside Cursor, it would be ideal to use authentic data representing how they actually use the app. User data reveals the behaviours executed and the workflows being deployed inside an app, which can form important signals for a model designed to support such activity and better align it with user intent. The technical paper for Composer 2 mentions the use of a “large code-dominated data mix” for pretraining and carrying out reinforcement learning with large datasets of coding tasks run in “environments that emulate real Cursor sessions as closely as possible.”

Federico Cassano, a researcher at Cursor, also somewhat admitted this in a podcast interview with Sequoia in which he agreed with the notion that Cursor “sits in the middle of so many interesting coding tokens [it] actually pretty uniquely have access to data to be to train at almost pretraining scale”:

What Cursor is doing here could be seen as natural evolution for AI companies for three main reasons:

1. Being reliant of frontier developers for the intelligence engines behind their apps makes their stack more vulnerable. Just look at the Fable fiasco with the US government last month.

2. Trying to engineer existing models, whether through clever system prompts or elaborate UIs, only gets you so far. Such engineering does not fundamentally change the weights of these models and therefore does not enable them to reliably bake into the model the nuances of how its app works and how users utilise it.

3. Model providers themselves have been making their own moves into the application layer. Anthropic’s Claude Code, OpenAI’s Codex and Gemini’s Antigravity are agentic coding products that compete directly with Cursor.

Consequently, it could be argued that AI companies should be looking to build their own models as a way to distance and protect themselves from frontier labs and establish deeper moats that make them more sustainable businesses in the long-run.

That being said, the influence of larger companies in the industry is never far away. Cursor explains in its blog post for Composer 2.5 that its partnership with SpaceX announced in April 2026 will provide access to 10x more compute to train its model in the form of millions of GPUs from Colossus, xAI’s data center located in Memphis, Tennessee.

The obvious consequence of building proprietary foundation models is that these AI companies move themselves to a different part of the supply chain. Instead of sitting firmly in the application layer of the AI ecosystem, building models also spreads them in the model layer.

And this move has consequences from a legal perspective that might easily be overlooked.

One relates to privacy and the rules around if and how AI companies can repurpose user data for model development. I have written about some of the issues here previously in relation to the GDPR:

But other implications come from the EU AI Act. This legislation includes specific provisions relating to general-purpose AI models and systems, with a complex set of rules applying to the builders and users of such models.

These provisions came into force in August 2025, and the Act will be enforceable in August 2026. It will therefore be interesting to know if companies have the Act on their radar when making these moves into the model layer, especially for those that operate on a global scale.

Addressing these legal implications early strengthens the moat for companies like Cursor. If they can show users and enterprises that their models are not only purpose-built and highly capable, but also compliant with EU law, the product becomes far more trustworthy and far more attractive and also ticks off an item that investors may have on their list.

In this newsletter I cover:

• How Composer 2 was developed

• The GPAI provisions in the AI Act

• How the GPAI provisions apply to Cursor’s development of Composer

• Whether the recent Omnibus changes anything

• What other AI companies looking to build their own models and serve EU users should be thinking about

Read more

There is certainly a divide when it comes to technology.

On one side you have people who are strongly averse to most of it. These people believe that technological development is ultimately bad for society and needs to be tamed. They believe in the ability of the state to protect people from the excesses of the today’s tech leviathans.

On the other side you have those who believe technological development is the answer to all or most of society’s problems. Just keep building and eventually the ends will justify the means.

And then there is something in the middle. Tech is great in many ways, but also has its downsides, and we should therefore focus on minimising the bad and maximising the good.

In the UK, this min-max middle is being squeezed out by policies that are aggressively anti-tech in a way that betrays the classic idea (or lazy stereotype) that the British people are somehow calm, restrained and reasonable.

In June, the UK government announced that it would be banning under 16s from accessing social media platforms. This would, according to the outgoing Keir Starmer, “give kids their childhood back” and “set a new normal for future generations.”

Obviously these are pretty loaded statements, but they are consistent with some of the other anti-tech content I have seen floating around. It’s not hard to find commentary these days about how screen time is frying your mind, shortening your attention span or causing brain rot, among other things.

And as part of that rhetoric, social media is so often cast as the digital fast food, something inherently dangerous that should be consumed with extreme caution if at all. Because if you do use it, you apparently risk sacrificing your childhood and inadvertently discarding a promising future by getting pulled into the dreaded black holes of internet brain rot and slop.

But as is often the case, the reality is more complicated.

Read more

There is a perception that the impact of AI on knowledge work will be pretty dreary for us.

AI makes expertise more available than ever before. It makes it more accessible (you can just prompt LLMs using natural language), as well as cheaper and faster to produce than humans are capable of.

This means that the production of stuff inevitably increases. Whe…

Read more

OpenAI is gradually rolling out ads to ChatGPT.

The news of ads being displayed inside its AI product emerged in November last year. The pilot started with logged-in US users in January 2026 on the Free and Go plans for ChatGPT. And more recently in May, the company announced that it would be expanding the pilot to users in the UK, Mexico, Brazil, Japan and South Korea.

But I decided to take a look at the small print to see how OpenAI is looking to make this lawful under the GDPR and what it means for users and AI products in the EU more broadly.

This is important because the terms of the bargain are being set quietly, in privacy policies most people never read, and they will shape what millions of users get in return for one of the most-used tools. Knowing that ChatGPT is becoming ad-supported means that your prompts, interests and conversation history will become targeting signals, and that the “free” tier now has a cost paid in data.

Read more

I like Aesop’s fables because they provide simple and accessible ways of conveying interesting themes about human behaviour.

Take the fox and the lion, which goes like this:

A very young Fox, who had never before seen a Lion, happened to meet one in the forest. A single look was enough to send the Fox off at top speed for the nearest hiding place.

The seco…

Read more

AI development can be seen as a sort of flywheel.

Cumulative improvements takes place at different parts of the AI system development lifecycle: pre-training, post-training, inference and agents.

With pre-training, it has been observed that if you increase the size of the model, the amount of data the model is trained or the amount of compute used for training, the performance of the model also increases. It is based on an observed strong correlation between these different aspects of the training operation and model performance.

Post-training focuses on techniques to improve the capabilities of the base model produced after pre-training is complete. Think reinforcement learning, fine-tuning etc.

Scaling laws have also been found to exist at inference. With this, the more time a model spends processing the input (i.e., ‘thinking’ or ‘reasoning’ about an answer to a prompt), the better the response it produces.

And then agents are models forming part of wider systems built to perform certain kinds of tasks in a given domain. They interact with the wider world through the use of tools, data and feedback loops to complete tasks with little human supervision.

All of these things link together to form a flywheel for AI:

1. Improved pre-training leads to improved base models

2. Improved post-training leads to improved foundation models

3. Improved reasoning leads to improved agentic capabilities

4. Deployed agents generate new data that is used to train future models

If this flywheel is allowed to persist, then maybe we get great leaps in AI capabilities.

But the reality is that this flywheel will not always be spinning. Or at least it will not be spinning as quickly.

I think a lot of people in the ‘AI will disrupt everything’ crowd forget about the real frictions and constraints that cannot always be engineered around and will make AI’s diffusion across the economy much slower and less disruptive than they predict.

The world is far less predictable and far more complicated than people realise. Anyone is capable of overfitting their perception of the world to their own experiences, but what they experience is merely a fraction of what happens in the world.

So the AI flywheel is limited by real constraints that should not be forgotten, and you can roughly fit them into three buckets:

1. The limitations of the technology

2. The scarcity of resources

3. The inevitability of governance

Read more

I have written previously about how AI will impact knowledge work. Simply put, AI makes the generation of content and knowledge much quicker and cheaper such that there may be less reliance on humans to do this sort of work.

But what AI does not change is the importance of taste and judgment - even if you can produce a document quickly, you still need to know whether that document contains everything it needs to contain and what you need to do with it.

In other words, you still need humans with subject-matter expertise to direct, verify and implement the outputs generated by AI systems.

I wanted to demonstrate this with a small test on some of the popular AI systems. Given that I work in tech governance, the test consisted of providing fictional scenarios with data protection issues and asking the AI system to address those issues.

ChatGPT was the first model I tested. I used the free version (though I am not sure what specific model this uses, this is not really apparent anymore, so it was probably some variant of GPT-5), with memory switched off and I also created a new chat for each scenario.

The main takeaway from this exercise is that AI is quite good at some things, but not everything. And if you use the general-purpose, out-of-the-box systems like ChatGPT for tasks like legal research, do so at your own risk.

Read more

There are some developers right now that think launching products in the EU and UK is impossible.

They look at the data laws in these regions and decide that they are too difficult to parse and follow, and so don’t even bother trying.

But this is a mistake.

You are missing out on big markets. That is over half a billion people you are skipping over and a h…

Read more

Generative AI is has taken a few forms to this point.

We first had the chatbot interface, which today still remains quite dominant. It is the one that people tend to be most familiar with, cementing its place with the release of ChatGPT in November 2022.

We then had auto-completion, particularly with coding. This made interaction with the model more indirect; you simply start typing your code in the editor and the model suggests what the rest of the code should look like.

Agents take this ambience further. With agentic architectures, models become the engines that power a wider system that operates autonomously usually with little human direction.

AI agents are designed to go beyond simple single-turn, input-then-output interactions. They can identify the steps required a complete task, access the right tools and data, and use feedback loops to verify that it has completed the task set.

Why are we moving more toward these agentic architectures? Because the chatbot form factor is difficult for most people to use. They do not always know what questions to ask and often struggle to unlock the full capabilities of the model sitting behind the text box.

Agents close this capability gap.

As I wrote previously:

[Agentic architectures] leverage AI capabilities to autonomously execute the right instructions, gather and use the relevant context, and complete actions on behalf the user. Agentic AI could bring into existence the kind of ambient digital assistants that help users navigate their digital world in a way that demonstrates the extent of the capabilities of these strange black boxes.

But for agents to work effectively, they need to be surrounded by other components. Agents are systems consisting of different parts that do not just instruct the model on what to do and how to do it, but also provide it with the resources to enable the completion of the task.

How those components are fitted together is crucial. As talks about in her newsletter on AI agents in knowledge work:

The broader point is that the debate is shifting from “which AI is smartest” to “which system can actually get work done inside an organization.” When intelligence is commoditized, the new scarcity is agency, and the ability to act across systems, with the right permissions, in the right sequence, with accountability attached. That is workflow. That is what the refineries are really selling.

So how do we build whole systems that can get work done?

Harness engineering.

Prompt engineering focuses on crafting detailed natural language instructions for the task the model should complete. Context engineering focuses on providing models with the relevant information to complete tasks with.

Harness engineering is the practice of building the scaffolding around the model that enables it to go beyond just producing text. It is what gives the model the ability to browse the web, run code, read files, remember what happened in previous sessions and take multiple steps toward a goal.

But harness engineering is not just about making models more capable. It is also about making models safer and more reliable.

And this is where AI governance comes in.

Governance work typically lives on paper: policies, inventories and risk assessments scattered across Word documents, spreadsheets and slides.

But this paperwork means nothing if it is not accompanied by methods for its practical implementation. Governance stays stuck as just words on paper.

Harness engineering is what transforms governance from mere paperwork into executable rules for AI agents.

It becomes a way to achieve governance-by-design, producing the safe and reliable AI agents that users and enterprises are ultimately looking for.

In this post, I break down what harness engineering entails and how it can be used to implement three core aspects of AI governance.

Read more

I saw something pretty mind-blowing the other week.

Whilst doomscrolling through some YouTube shorts, I was presented with video in the all-too-familiar street interview style that you may have seen before. This one featured the interviewer asking a guy whether he would want to know whether his girlfriend was actively using a dating app. The interviewer then started using an app called Cheater Buster AI to find the answer, asking some basic information and taking a photo of the girlfriend. You can watch the video for yourself:

Now it does seem like a lot of these videos are staged. Or this least seems to be the consensus when I searched for posts about this app on X. But I was still curious if this fakeness extended to the app itself - is there really a Cheater Buster app and, if so, how does it actually work? It seemed like a massive privacy red-flag.

And indeed there is such an app. On its ‘About Us‘ page on its website, Cheater Buster’s stated mission is to “give people a practical way to verify public profile activity and make informed decisions based on real signals instead of guesswork.”

The way these apps work is simple. You enter the name of the person you are searching for, as well as their approximate age and location, with the option to also upload a photo of the person. The app then compares the search query information with dating profiles on Tinder, returning a list of potential matches. The search results include profile pictures, bios and recent activity and change indicators (e.g., how long the person has been on Tinder for).

There is a notable lack of information available regarding how the app provides this service from a technical perspective. Tinder does not provide an official API, and the normal UX for its app only allows users to swipe through profiles presented to them by Tinder’s algorithms. There is no ability for users to directly search for other profiles. So how an app like Cheater Buster is able to retrieve dating profiles from Tinder is uncertain.

Also unclear is the facial recognition technology (FRT) being used for processing the images added to the search query. Computer vision techniques today are not only quite advanced but also widely available, with several different open-source implementations that can be used and integrated into other systems or apps. This has given rise to services like PimEyes, an image search engine that allows users to upload a person’s photo and find other images of that same person existing across the internet. Interestingly, Cheater Buster published a blog post in 2024 positioning its app as an alternative to PimEyes.

The reviews of Cheater Buster are mixed, with an average rating on Trustpilot of 2.9. Some people on Reddit claim that they have been able to find people using the app, but it seems that the quality of the matches is largely dependent on the quality of information provided in the search query.

What I find fascinating about this app is what it represents about privacy in our digital age.

There are two things in particular here:

1. Consumer surveillance tech continues to advance and proliferate

2. The general attitudes towards privacy and data protection still seem quite weak

Read more

There is something deeply concerning about the increased use of facial recognition technology (FRT) by private retailers in the UK.

Sainsbury’s has been trialing the use of this technology in its stores since September last year. Several other retailers have embarked on similar paths, including Asda, Sports Direct and Flannels.

Facial recognition is a com…

Read more

Privacy is going to have a renaissance in the age of AI.

Privacy and data protection will need to be built into AI apps and tools such that they become integral parts of the user experience.

They will become indispensable. Not add-ons or after-thoughts. Top priority requirements.

This view is not easy to appreciate because most people see privacy and data protection as relics of a distant past. As our digital world evolves, the concept of responsible and ethical use of people’s data falls further into a black hole of data maximisation, function creep and invisible processing.

Companies grab whatever data they can, use it for whatever purposes they want and all without the data subject’s knowledge.

This is the current state of affairs when it comes to privacy and data protection - nobody seems to really care.

But this is not entirely true.

In terms of the different attitudes towards privacy data and protection, you can broadly fit companies into three categories:

1. Those that care but struggle with it. They want to be compliant but find the frameworks burdensome, expensive, fragmented, and unpredictable.

2. Those that don’t care because they’ve made a rational calculation. They’ve looked at the expected cost of non-compliance (probability of enforcement multiplied by severity of penalty), compared it to the opportunity cost of diverting resources from product to compliance, and concluded that privacy and data protection is not worth prioritising yet. They’re not necessarily opposed to compliance; they just don’t think it’s material at their stage.

3. Those that don’t care because privacy and data protection have never been made to feel important. These companies haven’t even run the calculation of the second group. Compliance exists in a category of things they’ve filed under “boring, abstract, probably not relevant to me right now.” It’s homework. Nobody does homework voluntarily.

There are not many companies in the first group, there are slightly more in the second, but a lot are in the third group.

And you might think AI will only continue this trend. AI development has so far encouraged norms and practices that vanquish basic data protection principles, including purpose limitation, data minimisation and so on.

But this might be about to change.

Read more

I changed my mind on AI. And you probably should too.

When ChatGPT first came out and was blowing up, I wasn’t all that impressed. I tried it a few times and thought it was cool but did not see the value in talking to a chatbot apart from asking random useless questions.

I had come across these GPTs before. They came up during some research I was doing on how security and intelligence agencies were using AI for their operations. A 2020 report by the Royal United Services Institute detailed how AI could be used by such agencies for ‘cognitive automation’ in which AI can help analyse large volumes of data. It mentioned in this context OpenAI’s GPT-2 and how this model could be used for language analytics purposes, including for analysing transcriptions of captured audio. But not much else was written on the matter, and so I never looked into these language models any further at the time.

And then a short time later in 2022 when a wave of generative products came crashing in, including DALL-E 2, Midjourney and of course ChatGPT, I was wholly sceptical. Suddenly there was lots of talk about artificial general intelligence, AI safety, the prospect of AI taking all of our jobs and so on. But my understanding and interest remained fairly limited, and so I interpreted much of this as hype that should not be taken all that seriously.

And this is the attitude I had for a long time, barely using the tools whilst I stood to the side and criticised. I would read about AI but hardly use it myself.

It is the classic move for most people working in tech law and policy - dealing with the object of that to be governed at arms length and with only a minimal understanding of its mechanics. It is like the difference between static and dynamic analysis of an application. Legal professionals tend to be stuck constantly doing the former and therefore largely ignorant to how these AI systems work in real-world environments.

And there is little motivation to see it any other way. Legal experts are relied on for their risk-aversion and being able to see the worst case scenarios so that they can be mitigated against or avoided. Seeing AI this way is not particularly inspirational and certainly does not encourage one to actually want to test out these systems for themselves.

Eventually my passivity passed though and I decided that instead of sitting on the sidelines, I was going to get stuck in.

This is when I started experimenting with AI for tasks that I wanted to either offload or make much easier to do.

Among these experimentations included building an AI workflow for GDPR vendor reviews, a rather tedious and time-consuming task. I was able to work out the appropriate instructions and harnessing, with safeguards for reducing hallucinations whilst also making verification straightforward.

At this point I was no longer just talking about AI from a distance. I was enhancing my understanding through hands-on experience, something most legal professionals would not think to do.

The challenge I have ran into now, however, is not a technical one. Using AI is something I feel far more comfortable with now, and my experimentation with it will no doubt continue.

The challenge I now face is whether my increased use of AI is actually a good thing.

Read more

People often get distracted by the shiny new object.

The evidence that this is happening with AI is undeniable at this point.

CEOs and senior leaders get dazzled by the stories told by the frontier labs. Predictions are abound that these AI systems are so capable that they will replace all knowledge workers. New models or features get released and SaaS stocks tank. Some companies go as far as engaging in massive layoffs, thinking that large chunks of the workforce can be replaced by an army of cheaper AI agents.

AI has many advantageous of course. Despite being a legal professional and trained to see things with a more risk-averse eye, I can still see the various ways that AI can boost productivity, help make sense of unstructured information and uncover blindspots in my thinking.

But I can also see that AI does not solve all problems at once. It is not a silver bullet.

Like any other piece of technology, AI has its limitations. Hallucinations, getting stuck in recursive loops and all the quirks that come with operating large, non-deterministic systems programmed implicitly.

We are still in the very early days in terms of AI diffusion and adoption. Most people are not heavy users of AI, and therefore a minority know how to use it well.

And it is because of this that most people do not understand what AI is and what it is not.

It is not this magical machine that you can one-line prompt and it will immediately give you everything you need.

Yet some people basically have this perception of AI, probably because of the way it gets hyped by its developers. This mistaken perception leads to AI being treated as a solution looking for a problem.

The erroneous thinking here is something like this:

Because AI is apparently so amazing, and because everyone is using it, then surely I or our organisation can also use and benefit from it too?

This is a trap driven by FOMO and exacerbated by clouded judgment.

However, this is not just a case of misunderstanding the technology that AI is. I think there are other important aspects at play here.

Read more

It is becoming much easier for individuals to monitor each other at scale.

I remember a time when doorbell cameras were a rarity, and were even perhaps met with a natural reaction to mediums designed to capture data about you. Unnecessary, excessive, creepy.

But now as I walk through most neighbourhoods, doorbell cameras are everywhere. It is becoming rare to see a house that does not have one of these devices slapped onto or next to a front door, gawking at you as you walk past.

What has happened here? How have these devices, once frowned upon, now become almost an imperative for every household?

I think this trend reveals something interesting about how people view privacy.

Many people often acquiesce to convenience, despite the preferences, values or ethics that they might claim to hold. This is ultimately context-dependent, but it is fair to say that people’s boundaries are dynamic such that they may be willing to buy particular products or engage in certain activities which they may have previously disregarded.

We can observe across many different domains in our digital age. Social media is an obvious one, and AI chatbots are probably the next.

But there is something particularly intriguing that the rise of doorbell cameras reveal about the attitudes people have toward privacy, whether their own or that of others.

One norm that seems to be emerging is that if one places themselves in some sort of public realm or area, whether physical or digital, then the collection of their data is basically free game.

There is sometimes a perception that if a person has a public social media account, and posts content on that account that essentially anybody can view, the collection and use of that data is completely legitimate. If it is public, and anybody can see it, then surely anybody can use it? If you didn’t want people to use your content, then don’t post it in a public space.

But this misses an important point about data rights and their purpose. That data are public may mean that the level of privacy one can expect is reduced, but it is not zero and, in any case, the publicity of the data cannot be taken as some sort of implied consent for its endless use by others.

This applies in the physical domain too. Just because you record someone who happens to walk past your property does not mean you can do whatever you want with that recording.

The use of data ought to be dictated by the legitimacy and necessity of its use, as well as be used only for that purpose and accordingly limited in term of its nature and volume.

These principles I think are being respected by individuals less and less. We can talk about certain companies and the poor data practices that they follow. But at an individual level, I think the respect for privacy is getting weaker.

And maybe doorbell cameras are not the strongest example to demonstrate this with. There are some people who simply use them as a deterrent, and do not even bother to use the full range of their functionalities or might even admit if questioned that the damn thing doesn’t even work (or they don’t know how to work it).

So another perhaps more consequential piece of privacy-invasive hardware to cite here is recording glasses. A surveillance device dressed as a fashion accessory.

These are devices that enable a more ambient form of passive data collection. They can basically record anything you see in any location that you go.

The same emerging cultural dynamics are at play here: if you are in a public space where someone happens to be filming with their recording glasses, that is on you.

The thing about both doorbell cameras and recording glasses is that it takes away the agency of the surveilled in terms of whether their data are captured. They have no choice - if someone has one of these devices and you enter their periphery, then your data is collected. No opportunity to consent, to know what it might be used for or who it might be shared with. It is already in the hands of someone else.

And people may not even know that this is happening. The indicators that someone is wearing recording glasses are subtle and very difficult to spot at a distance.

Such devices contribute to a sort of fragmented surveillance at scale.

The question is whether data rights are built for this.

I think our legal frameworks were mostly built to deal with more centralised forms of surveillance, initially by states but then by companies. You regulated singular entities carrying out processing activities concerning lots of people; an easily identifiable bottleneck that can be addressed in one swoop.

But when the means of surveillance are fragmented across the population, giving individuals the ability to monitor lots of other people at scale, what recourse do those being surveilled have?

Share

Read more

The way a piece of technology is built and and how it works is always downstream of the incentives, opinions and choices of its builders and the context they are operating in. This is an obvious point that is often forgotten.

That technology is not neutral sometimes allows us to trace the impacts of technology back to the actions and thinking of its builders. Understanding the incentives, opinions and choices of these people is key for understanding why technology develops in a certain way or why it has certain impacts.

This backdrop is important when thinking about how the internet and digital products and services look today. Many of the products and services we have become accustomed to over the years are available for no upfront monetary cost for users. Search engines, social media platforms, and now AI chatbots all have free versions that essentially anyone with an internet connection can use.

But much of this free internet as we know it has been powered by ads. Google was one of the first companies to really work out how to build a business model that enabled its product to be free to use whilst still generating revenue. It figured out how to make revenue-generation essentially invisible to users yet still providing and updating a useful product.

And now OpenAI is doing the same by introducing ads to ChatGPT. The AI provider is currently rolling this out to ChatGPT free and Go users, with advertisers reportedly committing between $50,000 to $100,000 in ad spend.

The reasoning for implementing ads is similar to the constraints that Google and others had to face in the journey to profitability. However, I think the nature of its implementation for AI systems like ChatGPT will be different.

OpenAI’s pivot to ads is a sign of the reality they face - they need a reliable way to generate revenue. The company needs a way to cope with the immense increases in infrastructure spend on top of the high level of investments it has secured over the years. It would seem that subscriptions and enterprise deals are not sufficient.

Maybe OpenAI’s mission is still to build AGI eventually. But it may not survive to achieve this feat unless it develops a workable business model in the meantime to fund and power its efforts toward this grand milestone.

Google faced a similar problem in its early days. It bulked up its ad businesses due to pressure from investors and the need to somehow make money from its product.

When it comes to ads on Google, these are aligned to inferences made about users based on what the queries they enter into the search engine. On social media platforms, these inferences are derived from a wider range of user activity, including the people they follow and the content they engage with.

Ads in AI chatbots will be aligned to something different, something much deeper and higher-resolution than the data points that search engines or social media platforms can utilise.

AI ads will be aligned to a user’s actual thoughts, something very distinct from the search queries, social graphs and engagement metrics of yesteryears.

In this newsletter I go through this argument and explore:

• How AI systems sit closer to user intentions than any other system

• Why AI ads are inevitable

• What happens to data rights

Read more

The conflict between Anthropic and the US Government is a big deal, but not in the way that most people think.

It is not just a contract disagreement or an AI ethics debate. I think it represents something bigger.

To give a very simplified overview of what happened between Anthropic and the US Department of War:

• In mid‑2025, the Pentagon awards large multi‑year AI contracts (up to about 200 million dollars each) to several firms, including Anthropic, to help build out military AI capabilities

• As the government builds an internal generative‑AI ecosystem (often described as GenAI.mil), negotiations focus on how broadly military users can deploy Anthropic’s models

• The Department insists on “all lawful uses” language; Anthropic agrees except for two red‑lines: no mass domestic surveillance and fully autonomous lethal weapons - these safety guardrails become the core of the dispute

• The administration publicly hardens its stance, promoting an “AI‑first” warfighting strategy and rebranding rhetoric around the Pentagon as the “Department of War,” signaling a more aggressive doctrine

• In late February 2026, defense officials issue Anthropic an ultimatum: remove the contractual bans on mass domestic surveillance and autonomous lethal weapons or risk severe consequences

• Anthropic refuses before the deadline, reiterating that those uses are outside what its models can safely or ethically support

• After talks break down, the administration orders federal agencies to stop using Anthropic’s AI tools and begin transitioning away over several months

• The Department of War designates Anthropic a “supply chain risk to national security,” a label normally associated with foreign adversaries; this effectively bars defense contractors from using Anthropic products in government work

• Anthropic publicly rejects the designation as unlawful and politically motivated, arguing it is being punished for insisting on safety guardrails

• In March 2026, Anthropic files suit against the U.S. government, challenging the supply‑chain‑risk classification and seeking to protect its right to limit high‑risk military uses of its AI

I’ve read good takes on the situation by various writers here on Substack. You shoudl also check out the coverage from Privacat, Jasmine Sun and Dean W. Ball, just to name a few.

When reading all of this commentary, what came to mind was an overarching set of themes that could be threaded together.

In fact, the fight between Anthropic and USG felt quite familiar to me. My gateway into tech law and policy was state surveillance and the Snowden revelations in 2013. I became fascinated with the way technology and law collided together, a convergence of forces that advance and organise our societies.

This for me signalled one of the most important questions of our time (for which I have quoted Jamie Susskind, author of Future Politics: Living Together in a World Transformed by Tech, many times before because he puts it far me eloquently than I have):

To what extent should builders of powerful digital systems be concerned with data rights, and what does this look like in practice?

What I see here with Anthropic vs USG are three converging patterns:

1. The state depends on private tech firms more than the reverse, and this will only continue in the future

2. Safety guardrails for LLMs are somewhat of a technical illusion that creates a big governance gap

3. LLMs could bring surveillance, whether by the state or other entities, to a whole new level

In this newsletter, I cover these three conversing ideas (power inversion, safety guardrails and the evolution in surveillance) and how Anthropic vs USG is a glimpse at the defining tension of the 21st century.

Share

Read more

Many might be excited about adopting AI, or anxious that not doing so will leave them far behind the competition.

So they start ideating and brainstorming about how these systems could be used to improve internal processes or create better products for customers.

In doing so, some will have the foresight to think about the importance of governance - there should be a focus on building the thing the right way and not just building anything.

But then this is where organisations make the biggest mistake when starting their AI governance journey.

They start by drafting a policy.

Maybe they look for a template online. Or maybe they even use AI to generate one for them.

They draft it, put is somewhere on their intranet, and then that’s it. They think they have completed governance and now they can just focus on building or using their AI systems however they like.

Simple, right?

Well not really.

Policies are just words. They might represent the principles, rules, guardrails and standards you want to follow when building or using AI. But these things alone are simply not enough.

These things just represent intentions.

And while crystallising intentions in a formal document that everyone agrees to follow is important, starting your governance journey with a policy means that your policy will not be:

1. Aligned with how your organisation is using and/or developing AI systems

2. Backed up by practical measures that implement the policy

3. Actually known, understood and respected by staff

Without these things, your AI governance framework just never really gets going.

If you want to get AI governance right, do not start by writing a policy. You need to do other things alongside the drafting of the policy that will make your AI governance framework actually work.

You need to at least:

• Understand the state of play in your organisation

• Develop an AI strategy

• Build and implement the right measures (organisational, technical and legal)

• Establish feedback loops

In this newsletter, I go through each of these steps, including why they are needed and what they entail. If you like this content, make sure to share it with others who might also benefit.

Share

Read more

If you have been paying attention to the AI rhetoric these days, you will have likely heard a lot of people talk about ‘taste’.

I myself started seeing this concept being banded around for a while, and an increasing amount at the beginning of the year with all the craze around Claude Code.

Claude Code opens the door for everyone to build (almost) anything. And this opportunity is not just available to those who know how to code - you simply need natural language to get going. Explain what you want built, and the coding agent will do the rest.

But with AI taking away the expense and friction of execution, focus has now moved to the other parts of problem-solving with code. Because in a world where you can now build (almost) anything, what matters more than ever is building the right thing in the right way.

And this got me thinking; there seems to be such a thing as having good taste in AI development and deployment, which I do think is genuinely important. But what about having good taste when it comes AI governance? Is there even such a thing?

Initially, I thought this was a silly idea. Taste connotes creativity, imagination and novelty. These are not words that people think of when they think of governance. On the contrary they might instead think: onerous, tedious and a blocker.

However, the more I thought about it, the more I realised that taste in governance is not only real, but crucial.

Governance without taste fulfils the dreary perception that most have about the field.

Governance with taste fulfils legal requirements but also helps unlock the full potential of AI.

Tasteful governance melts the idea that legal professionals are mere luddites who stifle products and demonstrates a way to balance the engineering and lawyerly incentives.

What tasteful governance actually means

For me, taste is essentially a filtering mechanism.

Over time, you accumulate various bits of knowledge from the work you do. You build an understanding of the different problems that you or your clients face as well as the appropriate solutions to them.

These problems and corresponding solutions are compressed into retrievable principles and samples that can be applied to new sets of problems you encounter later on.

Taste is therefore a way of deciding how to sample from that knowledge base. It is about selecting the learned principles and samples from previous work that are most appropriate for the present work.

This is how taste works in other domains too, whether it be music, movies or people. Of all the music one has listened to, of all the movies one has watched, and of all the people one has interacted with, they know how to pick well by using a learned criteria that helps distinguish between the good, the decent and the bad.

Taste is therefore the ability to pattern-match, abstract and see the signal in the noise in various situations based on past learnings.

To apply this to governance, taste is not just about knowing what the law says. That is the easier bit really. You look up the correct rules for a client’s given situation and interpret their meaning accurately.

It is the next bit where taste is important - knowing how to implement the rules. And this task requires creativity, context-sensitivity and commercial awareness to not merely come up with a solution, but the solution. That is tasteful governance in a nutshell.

Read more