frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Can open-source prompt-injection detectors catch realistic AI agent attacks?

https://github.com/rudratoshs/buried-injections
6•northbridgedev•1h ago

Comments

verdverm•54m ago
Do these prompt injections work in the places this test placed them? (tool output)

I have my agents read other instruction files and they don't seem to get affected by the instructions found after a read/bash tool call. Curious if any analysis has been done to see if older prompt injection data sets are even effective anymore.

The whole thing looks heavily agent generated, my trust in them is not very high, how has this been validated or verified by a human?

Should we expect a magic solution in the near future? https://github.com/rudratoshs/taintgate

(side note, it seems my 'no emoji' system prompt line works really well, I forget how obsessed they can be with emojis)

I'm personally setting up to instead use a policy tuned agent on the tool calls themselves (rather than the output), so it never gets run if it has things that it shouldn't be doing. Mainly because they insist on working around instructions that say "don't" or permissions that restrict tools (eg: "git push": "deny" - where they just put the command in a script and run it there, bypassing hard checks)

hugopuybareau•31m ago
Very cool. Did you try any majority vote or some other kind of technique to combine several of them and maybe achieve better results ?

Early rogue AI agent activity and attempts to hack found on urlquery.net

https://transluce.org/agent-activity
59•snikolaev•2h ago•34 comments

Linux support is coming to Snapdragon X2 Series

https://www.qualcomm.com/news/onq/2026/09/snapdragon-summit-agentic-ai-pcs-linux
359•aaronday•9h ago•154 comments

Claude discovers a novel enzyme system with CRISPR-like repeats

https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
626•raahelb•13h ago•644 comments

ArXiv receives multiyear commitments to support it as an independent nonprofit

https://blog.arxiv.org/2026/09/23/arxiv-receives-multiyear-investment/
157•JohnHammersley•9h ago•20 comments

Ideas on modernizing the open-source desktop

https://lwn.net/SubscriberLink/1095425/2d9f411252325784/
87•signa11•5h ago•56 comments

Contrastive Language Models

https://contrastive-lm.notion.site/
47•erichocean•3h ago•7 comments

VSCode's SSH Agent Is Bananas (2025)

https://fly.io/blog/vscode-ssh-wtf/
211•Rapzid•11h ago•133 comments

Virtio-nvgpu: Near-native Nvidia GPU access inside a KVM guest

https://github.com/nestrilabs/virtio-nvgpu
79•WanjohiRyan•7h ago•40 comments

The "Windows XP Box" (2003)

https://www.mini-itx.com/projects/windowsxpbox/
133•doubletwoyou•2d ago•23 comments

Making portable my unportable transputer C compiler

https://nanochess.org/transputer_c_compiler.html
38•nanochess•2d ago•4 comments

Fixing the Portobello Police Station Clock

https://pointinthecloud.com/2026-04-11-211700.html
433•avidly•16h ago•101 comments

Meta VR Glasses

https://www.meta.com/vr-glasses/
374•polymorph1sm•8h ago•326 comments

Why 'What's Opera, Doc?' looks like that

https://animationobsessive.substack.com/p/why-whats-opera-doc-looks-like-that
42•CharlesW•16h ago•9 comments

Mercury 2.5 LLM hits 770 tokens per second

https://artificialanalysis.ai/models/mercury-2-5
97•Retro_Dev•9h ago•59 comments

Making Tailscale Faster

https://tailscale.com/blog/making-tailscale-faster
156•yarapavan•14h ago•59 comments

Women Who Sold Books Door to Door

https://daily.jstor.org/the-women-who-sold-books-door-to-door/
11•herbertl•2d ago•0 comments

Italian parliament votes for return to nuclear energy

https://apnews.com/article/italy-nuclear-chernobyl-4891b6b7c7791ae84db6b0bf0f7cf567
748•geox•15h ago•522 comments

The backend that ran every game we made, ten years and counting

https://www.m2h.nl/writing/the-toolkit-behind-ten-years-of-games/
4•MikeHer•1d ago•0 comments

The mystery animal on an ancient god's head

https://signoregalilei.com/2026/09/13/the-mystery-animal-on-an-ancient-gods-head/
98•surprisetalk•1d ago•27 comments

Six-year-old breaks women's world Rubik's Cube record [video]

https://www.youtube.com/watch?v=UCMRgvyTm08
43•1659447091•2h ago•17 comments

Tokens too cheap to meter

https://jyn.dev/tokens-too-cheap-to-meter/
291•teoruiz•22h ago•197 comments

A brief history of Windows scroll bar shortcuts

https://devblogs.microsoft.com/oldnewthing/20260922-00/?p=112719/
134•tybulewicz•14h ago•75 comments

Lambda MicroEgg

https://www.philipzucker.com/lambda_miller_egg/
39•philzook•3d ago•1 comments

Show HN: An open-source manufacturing ERP/MES/QMS

https://carbon.ms/self-hosted
29•barbinbrad•7h ago•14 comments

Data liberation: Apache Kafka's native cluster mirroring

https://developers.redhat.com/articles/2026/09/22/data-liberation-apache-kafka-native-cluster-mir...
11•fvaleri•1d ago•1 comments

LensVLM: Compressing long context as images, expanding only relevant pages

https://huggingface.co/apple/LensVLM-9B
74•victormustar•13h ago•7 comments

Gemini 3.8 text-to-speech

https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/
300•swolpers•16h ago•134 comments

Feds Target AI Critics as "Foreign Agents"

https://www.kenklippenstein.com/p/feds-think-ai-critics-are-foreign
274•nmeagent•7h ago•270 comments

Microsoft tried to ban "Microslop", and six months later it has given up

https://www.windowslatest.com/2026/09/24/microsoft-tried-to-ban-microslop-and-six-months-later-it...
3•theanonymousone•9m ago•0 comments

Radicle: Disclosure of Vulnerability in the Network Protocol

https://radicle.dev/2026/09/23/disclosure-of-vulnerability-in-network-protocol
140•lostmsu•16h ago•54 comments