privacy

Vitalik

2026-10-04 00:12:33

Vitalik says he is testing local AI, zkAPI and Tor to get personalized health advice

Ethereum co-founder Vitalik said in a post on X that he is running a self-experiment aimed at getting personalized diet and exercise recommendations from advanced AI models without exposing private information to remote systems. The setup uses a local Qwen 3.8 Flash Next model as the coordinator, while more powerful remote models are called as tools. To reduce privacy leakage, the design relies on the local model to draft queries, zkAPI to separate payment identity, and Tor to isolate network and IP information. Vitalik said the system is already working and has returned recommendations, but he also outlined several trade-offs. According to his post, Tor is not well suited for unlinking requests one by one, and latency can run 10x to 100x above the ideal level. He added that the current request-construction strategy is still not good enough. Qwen 3.8 Flash Next is currently running at about 20 to 30 TPS, while speeds above 100 TPS would make the setup feel meaningfully faster. He also noted a basic constraint of the approach: the more carefully users limit what remote models can see, the less help those remote models can provide.

Vitalik Buter

2026-10-04 01:35:12

Vitalik tests local-model orchestration for remote AI with a three-layer privacy setup

Vitalik Buterin said he is running a personal experiment that uses health and travel data to generate diet and exercise recommendations while trying not to expose private information to frontier AI models. In the setup he described, a local model, Qwen 3.8 Flash Next, acts as the orchestrator, while remote large language models are used as tools to make up for weaker reasoning and knowledge on the local side. The privacy design has three layers: the local model rewrites queries to reduce identity and writing-style leakage, zkAPI is used so payment rails do not reveal identity, and Tor is used to hide network metadata and IP addresses. Buterin said he has already received recommendations and that remote models improved the output. He also pointed to several shortcomings in the current design, including Tor being a poor fit for linking requests, privacy that may still fall short, latency running 10x to 100x above what it should be, suboptimal skill-file strategy, local-model throughput of roughly 20 to 30 TPS, and a trade-off where giving remote models less data also reduces how much they can help.

Vitalik

2026-10-04 00:29:43

Vitalik says he is testing a privacy-preserving AI setup using local and remote models

Vitalik.eth said in a Farcaster post that he is running a personal experiment that combines a local model with a remote frontier model to generate personalized diet and exercise suggestions from personal health and travel data while keeping privacy protections in place. According to the post, the system uses a three-layer privacy design. At the identity layer, a local Qwen 3.8B model builds the query request on the user’s behalf so writing style does not reveal identity. At the payment layer, zkAPI is used to hide payment information. At the network layer, Tor is used to conceal the IP address. Vitalik said the system is already working end to end, but he also listed three shortcomings: Tor is inefficient and introduces high latency for per-request unlinkability, the local model currently runs at only 20-30 TPS while a smooth experience would require 100+ TPS, and tighter data protection reduces how much help the remote model can provide. The related code has been submitted to Ethereum’s zkAPI repository.

NEAR

2026-10-03 06:58:49

NEAR co-founder Illia lays out ecosystem priorities with AI and privacy applications at the center

NEAR co-founder Illia, known on X as root.near, said on Oct. 3 that a wave of both new and returning developers has been building in the NEAR ecosystem over the past few weeks while development tools and workflows have become easier to use. In his view, the bigger question for developers now is not how to build, but what to build and how to reach users. Illia listed a wide range of product ideas. They include an AI Tamagotchi powered by NEAR AI, where on-chain prompts enable verifiable inference and each character exists as an NFT tied to NEAR for inference payments, with game elements such as feeding, customization and battles. He also described a universal payment component that would let merchants and apps add a checkout button so users can pay with any wallet and on-chain asset, with Shopify named as one potential commerce setting. Other ideas covered an AI encyclopedia combined with prediction markets, a private OTC market built on NEAR Intents, a tool dubbed Never get liquidated to automatically manage lending and perpetual positions, privacy-focused enterprise AI products, delta-neutral strategy tools, and products for private trading and social copy trading. Illia added that developers can seek support through NEAR Legion, make apps compatible with Aurora Intent Connect, and use community partnerships, Product Hunt and X for distribution.

digital footp

2026-10-03 02:31:37

PCMag guide says reducing your digital footprint starts with old accounts and data broker opt-outs

PCMag has published a practical guide on how people can reduce their digital footprint and limit the amount of personal data exposed on the public internet. The guide says the first step is to identify and manually close forgotten subscriptions and rarely used social media accounts, while backing up personal data before permanently deleting any profile. It also points readers to Google’s three-dot search result menu for privacy or legal removal requests, and notes that copyright holders can file DMCA takedown requests. Beyond search results, the guide recommends clearing browser cookies and site data on a regular basis, sharing less personal information when signing up for new accounts, rejecting non-essential cookies, limiting website permissions such as location and camera access, using HTTPS connections, and setting social accounts to private when needed. PCMag also highlights the role of data brokers, which collect and sell personal information from public records and other sources. Users can contact sites directly or use services such as Incogni, DeleteMe, and Optery to send opt-out requests, though the report says these services do not permanently erase personal data because brokers keep refreshing their databases. In California, the guide notes, residents can use the state’s free DROP platform to submit deletion requests to more than 600 brokers.

Apple

2026-10-02 18:18:02

Apple to tighten macOS full disk access controls over AI agent risks

Apple said it plans to add new controls around Full Disk Access on macOS, citing rising risks tied to increasingly capable AI agents. The company warned that broad access to user files, messages, email and browsing history could become more dangerous as AI agents gain stronger abilities. Apple said the move is intended to address these new risks. The update was cited by Techub, which referenced TechCrunch.

Meta disputes claim that Muse AI read Mac messages without a user’s consent

Meta is pushing back against allegations that its Muse AI agent accessed a user’s Messages on Mac without permission, saying the feature works only when users explicitly enable it. Andy Stone, Meta’s head of communications, said on X that the Messages integration in Muse for Mac is "entirely opt-in" and requires two settings to be turned on: Apple’s Full Disk Access in macOS and Muse’s own Messages connector. He added that both permissions can be revoked at any time. The dispute follows an account reported by AppleInsider and described by Inc. columnist Jason Aten, who said he denied message access during setup and later confirmed that Full Disk Access was off on his Mac mini. Even so, Aten said Muse suggested a column idea based on a private text exchange with his podcast co-host and flagged a deadline message from his editor. After asking how the app knew that information, he said Muse incorrectly claimed it had only seen notification banners, while he later found that his local Messages database had been synced. Meta Superintelligence Labs head David Singleton said that explanation from Muse was wrong and stated that the app syncs Messages data only when Messages access is enabled. He also said reading Messages requires three separate permission steps and that macOS protections cannot be bypassed, even if the app had a bug. The dispute lands as Amazon has already blocked Muse over separate concerns tied to browsing behavior and customer credentials.

Bitcoin

2026-10-01 00:57:35

Shielded Bitcoin proposal aims to add private transfers without changing Bitcoin

Bitcoin Magazine has highlighted a research proposal called Shielded Bitcoin, presented in a show featuring Misha Komorov, co-founder of Alloc Innit. The idea is to use zero-knowledge proofs to hide the sender, receiver, and amount in a Bitcoin transaction while avoiding changes to Bitcoin itself. According to the program description, the design does not rely on a soft fork, custodians, or bridges. The discussion covers how Bitcoin PIPEs could make the approach possible, along with open questions the proposal still needs to address. It also examines whether indexers and ZK rollups introduce trust assumptions, how Shielded Bitcoin compares with Monero and Zcash, what kinds of privacy it can and cannot protect, and the limits faced by early users when the privacy set is still small. Other topics in the episode include fees, block space, larger shielded transactions, and possible demand from users concerned about wrench attacks or from corporate treasuries. The show also touches on dark pools, governments, and the next wave of Bitcoin buyers. Bitcoin Magazine notes that Shielded Bitcoin remains a research proposal rather than a finished product. The post first appeared on Bitcoin Magazine and is credited to Patrick Green.