“It Is Fucked Up What Is Happening”: Signal’s Whittaker on AI in the Operating System
Trending Topics Jakob Steinschaden
Signal’s Meredith Whittaker says AI in phones and OSes is a privacy disaster. She thinks the real threat isn’t chatbots — it’s the operating system itself.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meredith Whittaker went to TechBBQ in Copenhagen and did not bother with polite language. The Signal president argued that the current AI boom didn’t appear out of nowhere. It grew out of the same advertising-and-surveillance model that has shaped big tech since the 1990s, and now that model is being pushed deeper into phones, laptops and other systems people use every day.
Her bigger point is that privacy has changed shape. It used to be about what companies could collect. Now it is about what models can infer. That matters because, in her telling, the technology stack has expanded from simple data gathering into a much wider machine for prediction, profiling and behavioral modeling.
Whittaker says the technical roots of today’s AI are older than the current hype suggests. Backpropagation has been around since the 1970s and 1980s, but it only became useful once companies like Google and Facebook had massive data sets and enough compute to feed them. In her view, that turning point in the early 2010s is what made modern AI viable, especially for engagement-driven products. The result is a feedback loop: more data makes better models, and better models create more demand for data.
She also sees a more immediate risk in how AI is being folded into operating systems. Her example was Siri’s summarization feature in iOS 27, which can take a screenshot of whatever is on screen and send it off-site for processing so it can summarize what the user sees. If Signal is open, that can mean a screenshot of a private message leaving the device, unless the user has blocked it. Whittaker says comparable behavior exists in Gemini and in ChatGPT’s integration with iMessage.
That is why she thinks Signal cannot just copy the AI playbook. The service is funded by donations, costs about 50 million dollars a year, and refuses to build products that depend on vacuuming up more user data. Its one AI feature is local face detection for blurring photos, because it keeps biometrics off a big tech server. For Whittaker, that difference is the whole point: privacy is not a slogan, it is the architecture.
My take — AI-written commentary, not fact-checked reporting
The ugly part here is that the industry keeps calling this convenience while quietly turning every operating system into a little surveillance clerk. Signal’s refusal to play that game looks less like purism and more like the only sane response left. If AI needs access to everything to be useful, then the product is not smart — it is nosy with better branding.
Read more about this at: Trending Topics