frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

The last line of defense must not be AI

https://worklifenotes.com/2026/06/12/the-last-line-of-defense-must-not-be-ai/
2•taleodor•1h ago

Comments

Someone•1h ago
FTA: “The final authority must sit behind a deterministic, non-bypassable gate. AI must never hold direct permissions for destructive, irreversible actions (deleting a production database, moving funds, pushing to prod). So the last line of defense must always be either human oversight or a deterministic script with no AI workarounds.”

That’s fine in theory, but won’t fly in practice for all destructive, irreversible actions. As an example, how do you prevent a chatbot from generating a highly insulting/racist remark or incorrect or illegal advice that will, later cost you millions?

Human oversight is (deemed) too expensive.

A deterministic script can detect known profanities, but may suffer from a variant of the Scunthorpe problem (https://en.wikipedia.org/wiki/Scunthorpe_problem), and won’t detect unknown profanities or creative ones that don’t use any words that are considered profane. A deterministic script also is very bad at detecting legal issues with responses.

“Don’t reply a chatbot” will work for that, but for many, that doesn’t seem to be an option.

taleodor•49m ago
It's not about that we should drop LLM completely from the mix, but something like AI -> LLM control -> old-school classifier control -> script / human oversight is the way. If something has potential to cause millions in damages, it should be subjected to human oversight (likelihood / impact analysis needs to happen early in the system design).

Vykar is a fast, encrypted, deduplicated backup tool written in Rust

https://vykar.borgbase.com
1•delduca•1m ago•0 comments

Google Sues to Stop Chinese Cybercrime Group from Using Its A.I

https://www.nytimes.com/2026/06/12/technology/google-lawsuit-china-ai-scams.html
1•ChrisArchitect•1m ago•1 comments

Open Knowledge Format

https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-d...
1•berlianta•2m ago•0 comments

Show HN: Memoriq – Private AI Memory for ChatGPT, Claude, Gemini and Grok

https://memoriq.me/
1•giekaton•6m ago•0 comments

WASI 0.3.0 Released

https://github.com/WebAssembly/WASI/releases/tag/v0.3.0
3•mavdol04•6m ago•0 comments

Claude Fable-5 Jailbreak

https://twitter.com/elder_plinius/status/2064776322979676227
3•vismit2000•7m ago•0 comments

Codex vs. Claude Code Desktop Apps

https://catalins.tech/codex-vs-claude-code-desktop-apps/
2•cmpit•7m ago•0 comments

Meta Down

17•reportinglurker•8m ago•10 comments

Show HN: Modern AWS SDK for Python

https://github.com/kap-sh/aws-sdk-python
2•karpetrosyan•8m ago•0 comments

Fable 5 burns tokens fast. I found an cost tracker for it

https://github.com/tigerless-labs/cost-xray
1•AllenH45•10m ago•1 comments

SpaceX's president is floating a Tesla merger as the company begins trading

https://qz.com/spacex-tesla-merger-gwynne-shotwell-ipo-061226
12•andsoitis•10m ago•3 comments

I'm delighted to rejoin the Sovereign Tech Fellowship

https://hugovk.dev/blog/2026/sovereign-tech-fellowship/
1•lumpa•10m ago•0 comments

Bernie Sanders' AI Sovereign Wealth Fund Plan

https://www.schneier.com/blog/archives/2026/06/bernie-sanders-ai-sovereign-wealth-fund-plan.html
2•speckx•11m ago•1 comments

Anthropic's Fable 5 one-shots pristine 3D graphics with Three.js

https://twitter.com/ChrissGPT/status/2065193150222663959
1•binyu•11m ago•0 comments

I am (not) a Failure: Lessons Learned From 6&1/2 Failed Startup Attempts (2025)

http://blog.rongarret.info/2025/01/i-am-not-failure-lessons-learned-from.html
1•mmarian•12m ago•0 comments

Big Bang Inside a Star: How a Gravastar Forms

https://www.uni-frankfurt.de/en/newsroom/meldungen/pressemitteilungen/2026/urknall-im-innern-eine...
1•layer8•13m ago•0 comments

Gram, a source code editor forked from Zed

https://codeberg.org/GramEditor/gram
1•marcuskaz•14m ago•0 comments

Sensemaking as the Heart of Expertise

https://commoncog.com/sensemaking-heart-of-expertise/
2•Tomte•15m ago•0 comments

Happo MCP: let your agent review your visual-regression and accessibility diffs

https://happo.io/blog/introducing-happo-mcp-server
2•lencioni•16m ago•0 comments

SpaceX IPO demand is approaching four times oversubscribed, source says

https://www.reuters.com/world/spacex-ipo-demand-is-approaching-four-times-oversubscribed-source-s...
2•Vaslo•17m ago•1 comments

European sunscreens are safer than American

https://www.ms.now/opinion/msnbc-opinion/sunscreen-united-states-fda-ingredients-rcna153526
5•qsi•17m ago•0 comments

Trajeckt: a fail-closed gateway that enforces what AI agents can do (~1.6ms)

https://traject.tamor.ai/
1•Bhuwan28•17m ago•0 comments

How Spammers Are Hiding Behind Google and the New York Times

https://www.comparitech.com/news/how-spammers-are-hiding-behind-google-and-the-new-york-times/
1•speckx•20m ago•0 comments

Guardian Runtime – Local firewall for AI coding agents and runaway costs

https://pypi.org/project/guardian-runtime/
6•Prajwal_Hage•20m ago•0 comments

Fable 5 is Anthropic's most "honest" model

https://twitter.com/thisritchie/status/2065416823898820889
3•mritchie712•21m ago•2 comments

International Archives Week 2026: ArchivesForJustice: Rights, Memory and Futures

https://www.ica.org/international-archives-week/iaw2026/
1•rbanffy•22m ago•0 comments

Georgia is about to have the biggest solar cell factory in US history

https://electrek.co/2026/06/11/georgia-is-about-to-have-the-biggest-solar-cell-factory-in-us-hist...
2•donohoe•22m ago•0 comments

Show HN: Quire – An ObsidianMD plugin for long-form writing

https://github.com/Dromena-xyz/quire
2•dromena•23m ago•0 comments

Show HN: RedNotebook AI open-source AI data notebook for Trino, +12 SQL engines

https://github.com/sanniheruwala/RedNotebookAI/blob/main/README.md
1•heruwala•23m ago•0 comments

Hunting the 30-Year-Old World of Xeen MT-32 Crash

http://finalpatch.github.io/xeen/
1•finalpatch•23m ago•0 comments