Every browser engine company builds on rented land. Vision-Language Ad Blocking doesn't.

#security #opensource #privacy #research

Every browser engine company builds on rented land. Vision-Language Ad Blocking doesn't.

Vision-Language Ad Blocking: DOM-Independent Content Filtering Using Multimodal LLMs

The Problem

The advent of Manifest V3 in Chromium-based browsers has systematically dismantled the technical foundations of traditional DOM- and request-filter-based ad-blocking extensions, which rely on WebRequest API interception and declarativeNetRequest rule sets . This paper presents a novel architectural departure: a vision-language model (VLM)-based ad-blocking system that operates independently of DOM structure, CSS selectors, or network request patterns.

What We Built

By leveraging the Qwen 2.5 VL 2B parameter model in its 4-bit quantized GGUF format , the proposed system performs real-time visual classification of rendered browser viewport regions to identify commercial advertisements, sponsored content, and dark-pattern UI elements. We demonstrate that visual inference achieves 94.3% precision in ad detection across 1,200 website samples, with a mean inference latency of 187ms on consumer-grade hardware .

The Research

This paper presents a novel architectural departure: a vision-language model (VLM)-based ad-blocking system that operates independently of DOM structure, CSS selectors, or network request patterns.

We demonstrate that visual inference achieves 94.3% precision in ad detection across 1,200 website samples, with a mean inference latency of 187ms on consumer-grade hardware .

This research demonstrates that sovereign, local-first AI infrastructure is not a future possibility ? it is a present reality.

Full citation: Alpasan, L.-K. (2026). Vision-Language Ad Blocking: DOM-Independent Content Filtering Using Multimodal LLMs. The Anticloud Research Corpus.

Read the full paper

Why The Anticloud

The cloud was supposed to liberate you from infrastructure management, but it delivered the opposite. It made you dependent on companies that monetize your data, lock you into their ecosystems, and change their pricing and terms at will. The Anticloud breaks that dependency entirely.

This is sovereign AI. Your inference runs on your machine, under your rules, without anyone else’s permission. The model answers to you, not to a corporation’s shareholders. It cannot be turned off remotely. It cannot be deprecated by a product manager. It cannot be changed without your consent.

Cloud is not a fallback mode in our architecture. It is not an option at all. The system was not designed to work offline with sync later — it was designed to work without ever being online. Connectivity is not a feature we support. It is a dependency we eliminated.

Every AI company today is actually a data company. They make their money from your usage, your prompts, your attention, your private information. We built the Anticloud so that model does not apply to you. We cannot monetize what we cannot access. We designed it that way on purpose.

There are no black boxes in the stack. Every component is open source. Every design decision is documented. Every claim we make about the system can be verified by running the code yourself. We do not ask for your trust. We give you the tools to verify.

You do not need permission from anyone to run AI on your own computer. The Anticloud makes sure that remains true.

The Anticloud requires one machine, one binary, and zero trust in anyone.

About the Author

My name is Lois-Kleinner Alpasan. I'm 23 years old. I built The Anticloud.

I started this because I looked at the AI industry and saw something wrong. Every major AI system requires you to send your data to someone else's server. Every "AI company" is actually a data company — they make money from your usage, your prompts, your files, your attention. They call it a service. I call it extraction.

I spent the last two years building an alternative. Not a feature, not a product, not a startup looking for an exit — an entirely different infrastructure stack. One where AI runs on your machine, for you, and never needs to phone home. One where privacy is not a feature you toggle in settings but a property of the architecture. One where you don't have to trust anyone because you can verify everything.

The project is near production-ready. Every component is open. Every claim is backed by published research. The code is documented. The ledger is verifiable. The binary fits on a laptop.

I'm not asking for trust. I'm asking you to read the paper, verify the claims, and decide for yourself whether the cloud is really necessary — or whether it was always just the default because no one bothered to build an alternative.

Follow the work: