frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Sheets Spreadsheets in Your Terminal

https://github.com/maaslalani/sheets
1•_____k•1m ago•0 comments

Archaeologists discover wreck of Danish warship sunk by Nelson 225 years ago

https://www.theguardian.com/science/2026/apr/02/archaeologists-discover-wreck-danish-warship-sunk...
1•zeristor•1m ago•0 comments

Show HN: Design to Code

https://www.absl.design/
1•absolute7•2m ago•0 comments

FDA had already warned the self-proclaimed 'fastest growing company in history'

https://www.drugdiscoverytrends.com/the-new-york-times-spotlighted-medvi-the-fda-had-already-warn...
1•thm•4m ago•0 comments

Emotion concepts and their function in a large language model

https://www.anthropic.com/research/emotion-concepts-function
2•dnw•4m ago•0 comments

Show HN: Cursor Cmd+K like and macOS spotlight like TUI for all terminals

https://github.com/64bit/commandOK
1•gigapotential•7m ago•0 comments

Anyone switch accounts for Claude Code, did you lose everything?

1•dpark2026•8m ago•0 comments

¡Haciendo Historia Celebrating PyCon US's First-Ever Spanish-Language Keynote

https://pycon.blogspot.com/2026/04/haciendo-historia-celebrating-pycon-uss.html
2•lumpa•22m ago•0 comments

Auralo: An nice new Radio App check it out

https://testflight.apple.com/join/mEtdrzZ5
1•marc0janssen•27m ago•2 comments

Trump fires Pam Bondi as US Attorney General

https://www.reuters.com/world/trump-fires-pam-bondi-us-attorney-general-cnn-fox-2026-04-02/
2•mgh2•31m ago•3 comments

AgentShift–One command migrates your OpenClaw agents to NemoClaw

https://agentshift.sh/
1•ogkranthi•39m ago•0 comments

Uber engineer alleges hostile 'boys club' culture, firing after cancer leave

https://archive.org/details/10127280
3•nickvec•42m ago•0 comments

Spath and Splan

https://sumato.ai/posts/2026-04-04-spath-and-splan.html
1•jasonmoo•43m ago•0 comments

Ask HN: Interactive Car Mechanics Guide?

1•id00•47m ago•0 comments

Scientists are working on "everything vaccines"

https://economist.com/science-and-technology/2026/04/01/scientists-are-working-on-everything-vacc...
5•andsoitis•47m ago•1 comments

Donald Knuth: Open Letter to Condoleezza Rice (2002)

https://www-cs-faculty.stanford.edu/~knuth/rice.html
3•car•48m ago•1 comments

100 Years of the Iron Ring

https://engineerscanada.ca/news-and-events/news/100-years-of-the-iron-ring-a-symbol-of-an-enginee...
1•jruohonen•54m ago•0 comments

Vibe coded a design tool for a client handover as a non-technical founder

https://www.ugh.design
2•jayantrao94•59m ago•1 comments

Video Friday: Digit Learns to Dance

https://spectrum.ieee.org/video-humanoid-dancing
2•jruohonen•1h ago•0 comments

AdGuard ad trackers What ad-based surveillance does to your traffic

https://adguard.com/en/blog/adguard-ad-tracker-report-2025.html
2•XzetaU8•1h ago•0 comments

Pale Blue Dot

https://en.wikipedia.org/wiki/Pale_Blue_Dot
3•thunderbong•1h ago•1 comments

EU cyber agency attributes major data breach to TeamPCP hacking group

https://therecord.media/european-commission-cyberattack-teampcp
4•jruohonen•1h ago•1 comments

Show HN: AI Dev Board – Job Board for AI Developers with a Full REST API

https://aidevboard.com/
2•8bitconcepts•1h ago•0 comments

Ask HN: Why still embed heavy 3rd-party iFrames for simple social proof?

2•LordKode•1h ago•0 comments

Show HN: HyprMac – I missed Hyprland after switching to Mac, so I built it

https://github.com/zacharytgray/HyprMac
2•zachtgray•1h ago•0 comments

Thoughts on AI and Research [pdf]

https://economics.mit.edu/sites/default/files/2026-04/IA%20AI%20note_1.pdf
2•jxmorris12•1h ago•0 comments

Jungle old school drum and bass radio

https://radio.aklein.studio/public/lounge24_radio
3•misterthp•1h ago•0 comments

It's open season for refusing AI

https://www.bloodinthemachine.com/p/its-open-season-for-refusing-ai
7•HotGarbage•1h ago•2 comments

No luck for Broadcom as Netflix and Quinn Emanuel succeed in nullity claim

https://www.juve-patent.com/cases/no-luck-for-broadcom-as-netflix-and-quinn-emanuel-succeed-in-an...
2•breve•1h ago•0 comments

How to Evaluate Claude Skill Output Quality for Prompt-to-SQL Scenarios

https://dekart.xyz/blog/how-to-evaluate-claude-skill-output-quality-for-prompt-to-sql-scenarios/
2•delfrrr•1h ago•0 comments
Open in hackernews

A Big Alignment Loophole of Current Froniter LLMs

https://github.com/wuyoscar/ISC-Bench
3•pythonsen•1h ago

Comments

pythonsen•1h ago
[LIVE DEMO] AI Agents Jailbreak Themselves Without Any Attack

Normally, getting an LLM to produce harmful content — self-harm instructions, weapon tutorials, exploit code — requires a pretty sophisticated attack. Prompt injection, jailbreaks, adversarial suffixes, the whole arms race.

I found that in an agent setting, you don't need any of that. You just give the model a normal task — say, training a LlamaGuard content moderation model — and it will produce a full harmful text dataset on its own. No refusal. No hesitation. It thinks it's doing its job.

I tested 100 frontier models. Basically every model can be triggered this way. GPT-5.4, Gemini 3.1 Pro, Claude Opus 4.6 — all of them. Every major provider. Zero adversarial effort required.

This is a big deal for anyone deploying agents in production — tools like Openclaw, Claude Code, Codex, or any agentic framework that gives LLMs file access and code execution. If your agent touches sensitive data in science, healthcare, or security workflows, it could generate harmful content as a side effect of doing its job. I want to share this finding because I think both developers building on LLMs and normal users need to be aware. This is real — I've included live demos as proof so you can see it happening, not just take my word for it:

85 reproducible prompt if you want to try it yourself: https://github.com/wuyoscar/ISC-Bench