Skip to main content
Eye on AI

#339 Eamonn Maguire: Your Child Has a Data Profile Before They're Born

45 min episode · 2 min read
·
Eamonn Maguire

Episode

45 min

Read time

2 min

Topics

Investing, Fundraising & VC, Artificial Intelligence

AI-Generated Summary

Key Takeaways

  • Data profile construction: Companies like Google build behavioral profiles from as few as three data points — an Instagram signup, a newsletter subscription, and a search query — then use ad click patterns to fill profile gaps, inferring religion, politics, and income without users ever directly disclosing that information. Treat every signup as a data disclosure.
  • AI training data and trust: Free users on platforms like ChatGPT default to contributing their data for model training. Enterprise contracts offer opt-outs, but access remains. Anthropic faced a $1.5 billion lawsuit for purchasing and scanning thousands of books to extract training data. Users should assume any unencrypted input to a proprietary model is potentially retained.
  • Open model evaluation: Truly open AI models — including NVIDIA's Nematron series (20B, 120B, and upcoming 500B parameter versions), Allen Institute's OLMo, and Switzerland's Apertus — publish training data sources alongside weights and code. Proton's Lumo deploys these models on private infrastructure, rotating to frontier-performing options like Qwen and GLM as benchmarks shift.
  • Born Private and pre-birth data exposure: A child's data profile begins accumulating before birth when parents email fertility clinics or gynecologists through Gmail, triggering ad targeting for pediatricians and hospitals. Proton's Born Private program lets parents reserve a ProtonMail address for a child at birth via a $1 symbolic donation, anchoring the child's digital identity in an encrypted system from day one.
  • Privacy-preserving AI context: Proton's Lumo implements local document indexing through its Projects feature, allowing enterprise teams to link encrypted Drive folders. At query time, relevant documents are retrieved locally and injected into the prompt context before being sent to the GPU — meaning Proton's servers never read the content, yet the model still receives full business context for accurate responses.

What It Covers

Eamonn Maguire, Head of AI at Proton, explains how data profiles are built on individuals before birth through email metadata and behavioral tracking, and how Proton's Born Private initiative and encrypted ecosystem — including Lumo AI, ProtonMail, and Proton Workspace — aim to counter this surveillance infrastructure.

Key Questions Answered

  • Data profile construction: Companies like Google build behavioral profiles from as few as three data points — an Instagram signup, a newsletter subscription, and a search query — then use ad click patterns to fill profile gaps, inferring religion, politics, and income without users ever directly disclosing that information. Treat every signup as a data disclosure.
  • AI training data and trust: Free users on platforms like ChatGPT default to contributing their data for model training. Enterprise contracts offer opt-outs, but access remains. Anthropic faced a $1.5 billion lawsuit for purchasing and scanning thousands of books to extract training data. Users should assume any unencrypted input to a proprietary model is potentially retained.
  • Open model evaluation: Truly open AI models — including NVIDIA's Nematron series (20B, 120B, and upcoming 500B parameter versions), Allen Institute's OLMo, and Switzerland's Apertus — publish training data sources alongside weights and code. Proton's Lumo deploys these models on private infrastructure, rotating to frontier-performing options like Qwen and GLM as benchmarks shift.
  • Born Private and pre-birth data exposure: A child's data profile begins accumulating before birth when parents email fertility clinics or gynecologists through Gmail, triggering ad targeting for pediatricians and hospitals. Proton's Born Private program lets parents reserve a ProtonMail address for a child at birth via a $1 symbolic donation, anchoring the child's digital identity in an encrypted system from day one.
  • Privacy-preserving AI context: Proton's Lumo implements local document indexing through its Projects feature, allowing enterprise teams to link encrypted Drive folders. At query time, relevant documents are retrieved locally and injected into the prompt context before being sent to the GPU — meaning Proton's servers never read the content, yet the model still receives full business context for accurate responses.

Notable Moment

Maguire describes how Instagram's algorithm continuously served a 14-year-old named Molly Russell increasingly graphic content related to suicide, ultimately contributing to her death. A documentary by director Mark Silver examines how platform engagement optimization can reshape a child's psychology without any parental awareness.

Know someone who'd find this useful?

Episode Transcript

Can you introduce yourself and give that background? My background is varied, I would say. I started most of my professional type of career was in bioinformatics. It was basically using computational techniques to analyze, the genetic sequences or protein sequences and so on. And that very much got me into the academic mindset of things. And I went and did a master's in bioinformatics and then a PhD in computer science, Largely focused on also quite a bit of bioinformatics, but I also worked in security applications, a lot of stuff in insider threat, for example. Worked in things with digital humanities, also like analyzing poetry for the way things were are spoken and how it's how you can visualize that type of information. So I my my major interest for a lot of my life was working on visualization, data visualization. So the intersection of computer graphics, mathematics, statistics, machine learning later, I would say, and then the computer science aspect of it. And visualization opened up a lot of different avenues, in fact. So I worked in biology. I worked in security. I worked in digital humanities. I I ended up doing my postdoc at CERN. I spent a couple of years there, then I went and worked in finance for three years, taking a lot of what I learned from, I would say, the previous my previous career cycles into finance also. So I did a lot of work applying the techniques typically used in biology or bioinformatics, but doing it using those techniques in financial time series analysis, for example, or looking at different ways of correlating different signals together. And after that, I ended up working at Facebook where I worked a lot in more the there's more machine learning and data science for all external and internal threats. So everyone trying to get into Facebook's network, preventing that happening. Anyone trying to exfiltrate data out of Facebook from within Facebook, also stopping that as well. So more detection engineering. And then after that, I came to Proton. So I've been at Proton Life for six years. Okay. I gotta ask the the digital humanities is fascinating. What you're creating visualizations of language patterns and poetry. Is that is that essentially what you're doing? One project, in fact, was looking at how the tongue moves or how there's basically different ways. There's a representation about how you create sounds, like, from the back of the throat, from the front, from the tongue, and so on. And, basically, we had a grid representing that, and then we're able to visualize the changes of this over time. So you could see how different poems were supposed to be read in a more smooth way versus those which are more harsh tongues where you have big dynamic changes in positions, which manifested themselves then as being very large changes in in tonality, for example. So it was cool. I I like everything, which is interesting, and we …

Get the full transcript (7,856 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all Eye on AI transcripts →

You just read a 3-minute summary of a 42-minute episode.

Get Eye on AI summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links.

Tools

  • Lumo AIRecommendedBy guest

    by Proton

    Proton's Born Private initiative and encrypted ecosystem — including Lumo AI, ProtonMail, and Proton Workspace — aim to counter this surveillance infrastructure.
  • Truly open AI models — including NVIDIA's Nematron series (20B, 120B, and upcoming 500B parameter versions), Allen Institute's OLMo, and Switzerland's Apertus — publish training data sources alongside weights and code.
  • Proton WorkspaceRecommendedBy guest

    by Proton

    Proton's Born Private initiative and encrypted ecosystem — including Lumo AI, ProtonMail, and Proton Workspace — aim to counter this surveillance infrastructure.
  • ProtonMailRecommendedBy guest

    by Proton

    Proton's Born Private initiative and encrypted ecosystem — including Lumo AI, ProtonMail, and Proton Workspace — aim to counter this surveillance infrastructure.
  • Proton's Lumo deploys these models on private infrastructure, rotating to frontier-performing options like Qwen and GLM as benchmarks shift.
  • by Allen Institute

    Truly open AI models — including NVIDIA's Nematron series (20B, 120B, and upcoming 500B parameter versions), Allen Institute's OLMo, and Switzerland's Apertus — publish training data sources alongside weights and code.
  • by NVIDIA

    Truly open AI models — including NVIDIA's Nematron series (20B, 120B, and upcoming 500B parameter versions), Allen Institute's OLMo, and Switzerland's Apertus — publish training data sources alongside weights and code.
  • Born PrivateRecommendedBy guest

    by Proton

    Proton's Born Private program lets parents reserve a ProtonMail address for a child at birth via a $1 symbolic donation, anchoring the child's digital identity in an encrypted system from day one.

other

More from Eye on AI

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

Explore Related Topics

This podcast is featured in Best AI Podcasts (2026) — ranked and reviewed with AI summaries.

Read this week's Investing & Markets Podcast Insights — cross-podcast analysis updated weekly.

You're clearly into Eye on AI.

Every Monday, we deliver AI summaries of the latest episodes from Eye on AI and 192+ other podcasts. Free for one show.

Start My Monday Digest

No credit card · Unsubscribe anytime