privacy
Google's private AI runs on sealed hardware, not on encrypted math News
Google's shipping private inference product runs Gemini inside hardware enclaves on custom chips, which is confidential computing rather than homomorphic encryption, and the company's actual homomorphic work is an unsupported research compiler.
Encrypted inference: can a model answer a question it cannot read? Lesson
The two competing ways to run AI on data the server is not supposed to see: sealed hardware enclaves, which ship today and are fast, and homomorphic encryption, which is mathematically stronger and still far too slow.
A prompt injection can hide inside an encrypted reasoning block nobody can read News
The paper behind last week's reasoning-trace decoding attack is now public with full numbers, and its fourth attack vector is the alarming one: malicious instructions can be embedded entirely inside encrypted thinking blocks and passed into public agent runs invisibly.
Vision models can tell which camera took a photo, and they use it to cheat News
A paper accepted to ECCV 2026 shows that image encoders pick up invisible pixel-level traces left by cameras and image processing, then lean on those traces as a shortcut for the actual task, degrading when the equipment changes.
Models invent about 40 percent of what they claim to know about you News
A new study measuring 12 language models across more than 143,000 judged claims found every one of them invented or stereotyped between 35 and 49 percent of what it asserted about a user, and that the models most confident they were being careful were the ones fabricating most.
Differential privacy: a mathematical promise about what a model can reveal about you Lesson
Differential privacy is a formal guarantee that a system's output would look almost the same whether or not any single person's data was included, enforced by adding carefully calibrated noise. It is the only privacy protection for machine learning that comes with a provable bound rather than a hope.
A portable persona skill carried one user's verbal tics into unrelated conversations News
Researchers benchmarking persona skills, the reusable artifacts agents distill from your interaction history, found that agents equipped with one reproduced that person's distinctive phrasing in unrelated conversations up to 87.7 percent of the time, and that a watermarking defense meant to prove provenance detected nothing at all.
A judge did not rule that ChatGPT users have no rights to their chats News
A New York magistrate denied one individual permission to intervene in the OpenAI copyright litigation, and the order explicitly says the data preservation hold was for a possible spoliation inquiry rather than to hand conversations to the New York Times.
Training data deduplication: why the same text twice makes a model worse Lesson
Deduplication is the process of finding and removing repeated or near-repeated documents from a training corpus, and it reliably makes language models better, cheaper, and safer - the same text seen many times gets memorised rather than learned from.
Hugging Face ships The Stack v3, and this time the code is actually in the box News
Hugging Face released The Stack v3, a code dataset of roughly 4.9 trillion tokens across 173 million repositories with source text embedded inline, ending the v2 arrangement where researchers had to fetch the actual file contents themselves.
Cactus Ships a Phone-Sized Model That Knows When to Ask the Cloud, With a TLS Footgun News
Cactus released a Gemma-4 model with a tiny probe that scores how likely its own answer is wrong and routes uncertain queries to the cloud, but the cloud path ships full conversations and disables TLS verification by default.
Austria Starts Rolling Out GovGPT, a Sovereign AI Run on Its Own Data Centers News
Austria began deploying GovGPT to over 180,000 federal staff, processing all queries on national computing infrastructure with no third-party sharing, making sovereignty a matter of operating model rather than a single AI.
Model extraction attacks: stealing an AI through its own API Lesson
A model extraction attack tries to copy a machine-learning model you can only query, not download, by sending it many inputs and learning from its outputs. Depending on the goal, an attacker can clone the model's behavior, recover pieces of its internals, or reconstruct a rival model cheaply, which is exactly the fear driving today's AI 'distillation' disputes.
ChatGPT ads are live, and OpenAI has quietly built a full ad stack behind them News
OpenAI is running labeled sponsored cards below ChatGPT answers for free users in a US beta, and its own developer docs reveal a conventional ad-tech layer underneath, with a tracking pixel and a conversions API that bridge each ad click to a downstream purchase, even though the model's answers stay separate from advertisers.
Samsung Health asks users to let AI train on their health data - or lose cloud sync News
Samsung Health began prompting users to consent to AI training on intimate data including sleep, nutrition, medication and cycle tracking, with reports that declining costs cloud sync - a consent-under-duress pattern privacy law treats as invalid.
A researcher says xAI's coding tool uploads your whole repo -- secrets, unread files, and all News
An independent wire-level teardown found that xAI's Grok Build CLI uploads an entire code repository, including .env secrets and files the AI never read, to an xAI cloud bucket -- and the model-improvement opt-out does not stop it.
Machine unlearning: how to make an AI forget Lesson
Machine unlearning is the problem of removing specific knowledge - a person's data, a copyrighted book, a dangerous capability - from a trained model without retraining it from scratch and without damaging everything else it knows.
Claude Code Users Report Other People's Data Showing Up in Their Sessions News
Two new GitHub issues describe unexpected data appearing in Claude Code sessions, with one confirmed case of another user's live server credentials leaking in and being used without authorization.
Claude Code was quietly fingerprinting requests through a hidden mark in the date News
A reverse-engineer found that Claude Code secretly changes tiny characters in the date it sends the model - a covert marker aimed at spotting resellers and copycats.
What should an AI agent remember about you, and what leaks when it does? News
Researchers are asking whether AI agents are ready for real long-term memory, just as another study shows how much an agent's memory can quietly give away about the people it served.
What does your AI actually remember about you? News
Two new studies stop trusting that agent 'memory' works and start measuring it directly, with results that carry a privacy sting.
Agent memory: how an AI remembers you after the conversation ends Lesson
Why most AI assistants have amnesia, the difference between short-term context and real long-term memory, and why remembering you is both what makes agents useful and what makes them a privacy risk.
HEIR Tool
Google's compiler for fully homomorphic encryption: write a high-level program with annotations marking which values are secret, and it compiles down to backends including OpenFHE, Lattigo, tfhe-rs and Jaxite. Explicitly not an officially supported Google product.
Gemma-4 WebGPU Kernels Tool
A demo running Google's Gemma-4 model directly inside a web browser using your device's graphics hardware — private, on-device AI with no server and no data leaving your machine.
Am I in The Stack? Tool
Lets a developer check whether their GitHub repositories were included in The Stack code dataset, and points to BigCode's removal process. Opted-out repositories are dropped before each patch release.