I've been experimenting with Claude Code, ChatGPT Agent, and OpenClaw to perform more open-ended tasks for me online. A big blocker I've hit on shopping and research tasks is the agent getting a key piece of info wrong.

For example, in one case, my agent decided to add a brand I don't like to the cart because the site flagged it as almost sold out

The HN crowd is probably pretty aware of the threats and can avoid them while browsing. But what about their agents?

I tried prompting, but it was ineffective, because once the AI saw the threat, it polluted/distracted its context

Looking at the research, I came across a couple of papers, SusBench and Decepticon. The Deception research benchmarks indicate that increased reasoning can perform worse, because the model rationalizes the dark pattern

So it seems the best approach has to be removing the information before it can pollute/poison the context

In my day job, we have a browser extension that started as a productivity extension. However, contact centers started using us for neutralizing insider, fraud, and social engineering threats.

So my team set out to create a browser extension to neutralize all the threats AI agents face

We're focusing on open-ended tasks, because the best practice for routine tasks is to have the agent script repeat actions

It's also a tricky area since AI agents view the web in different ways: DOM, a11y tree, and visually. So we needed to account for those differences in how we detect and neutralize threats

The extension we created is agent-browser-shield, which defends against three primary threats:

- Prompt Injection - Dark Patterns - Context Pollution

It's free and source-available on GitHub, ClawHub, and the Chrome Web Store: https://github.com/pixiebrix/agent-browser-shield

We plan on making an enterprise version that pairs with our low-code engine for letting teams easily create custom rules for business-specific sites and internal tools

Looking forward to feedback! Especially curious if anyone has agent traces that got poisoned or sites to red team against!

Framework Laptop 13 Pro Deep Dive – LPCAMM2

AI costs how much? GitHub Copilot users react to new usage-based pricing system

I made a streaming feature where you can predict what happens during the stream

The SSRI Debate Flares Again

Ecdsa Fail Challenge

Show HN: Erebus, A VPN client that runs rootless in user space

Show HN: PersonalPod – Saving my long-distance relationship with a podcast

WebSSH: A modern, feature-rich web-based SSH terminal with SFTP file manager

My Coding Setup

New Texas Instruments 5532 chips are not the 5532's we've used for decades

Show HN: Nutrepedia – nutrition info in 29 locales built with Clojure and Htmx

GitLab cuts 14% of staff as it scales its platform to serve AI workloads

Tidal Wave: CachyOS Is Now Bigger Than Arch Linux Ever Was

DrakoFlow – A serverless, open-source text-to-diagram tool

Amazon-owned Ring should pay Americans for scanning their faces, lawsuit says

Show HN: Agent-browser-shield – free extension to protect AI agents on the web

Nvidia: Nemotron 3 Super

Today a crazy quantum story just got wilder

NPM-Scan: Detecting Supply Chain Attacks (100% Validation on Real Campaigns)

'Close to the Terminator narrative': the dawn of self-improving AI

AI Delegation Starts with Inspectable Work

Renowned Warhammer Artist, John Blanche, Passes Away

FIA Operating Systems FiaPhy Dtdss Algorithm

Resource consumption of AI: The insatiable industry and its costs

Thanks to Robots, Ukraine Is Now Talking About Winning, Not Just Surviving

Show HN: Mini-split BTU calculator and state-by-state heat pump rebate guide

Librarians on Horseback

Dreambeans, an app that connects you with what matters

Ghosts

What 100k concurrent sandboxes has taught us so far

I am developer in building distributed systems,how to get into Anthropic?