frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Microsoft 365 Copilot – Arbitrary Data Exfiltration via Mermaid Diagrams

https://www.adamlogue.com/microsoft-365-copilot-arbitrary-data-exfiltration-via-mermaid-diagrams-...
114•gnabgib•3h ago•18 comments

Show HN: MyraOS – My 32-bit operating system in C and ASM (Hack Club project)

https://github.com/dvir-biton/MyraOS
103•dvirbt•6h ago•9 comments

Sandhill cranes have adopted a Canada gosling

https://www.smithsonianmag.com/science-nature/these-sandhill-cranes-have-adopted-a-canadian-gosli...
35•NaOH•4d ago•7 comments

Are-we-fast-yet implementations in Oberon, C++, C, Pascal, Micron and Luon

https://github.com/rochus-keller/Are-we-fast-yet
31•luismedel•3h ago•7 comments

A definition of AGI

https://arxiv.org/abs/2510.18212
155•pegasus•8h ago•232 comments

You already have a Git server

https://maurycyz.com/misc/easy_git/
406•chmaynard•15h ago•325 comments

We Saved $500k per Year by Rolling Our Own "S3"

https://engineering.nanit.com/how-we-saved-500-000-per-year-by-rolling-our-own-s3-6caec1ee1143
52•mpweiher•5h ago•26 comments

Ken Thompson recalls Unix's rowdy, lock-picking origins

https://thenewstack.io/ken-thompson-recalls-unixs-rowdy-lock-picking-origins/
90•dxs•9h ago•6 comments

Sphere Computer – The Innovative 1970s Computer Company Everyone Forgot

https://sphere.computer/
19•ChrisArchitect•3d ago•2 comments

How I turned Zig into my favorite language to write network programs in

https://lalinsky.com/2025/10/26/zio-async-io-for-zig.html
23•0x1997•2h ago•3 comments

NORAD’s Cheyenne Mountain Combat Center, c.1966

https://flashbak.com/norad-cheyenne-mountain-combat-center-478804/
89•zdw•5d ago•44 comments

A bug that taught me more about PyTorch than years of using it

https://elanapearl.github.io/blog/2025/the-bug-that-taught-me-pytorch/
339•bblcla•3d ago•67 comments

Researchers demonstrate centimetre-level positioning using smartwatches

https://www.otago.ac.nz/news/newsroom/researchers-demonstrate-centimetre-level-positioning-using-...
24•geox•1w ago•7 comments

System.LongBool

https://docwiki.embarcadero.com/Libraries/Sydney/en/System.LongBool
36•surprisetalk•5d ago•29 comments

Show HN: Helium Browser for Android with extensions support, based on Vanadium

https://github.com/jqssun/android-helium-browser
26•jqssun•4h ago•8 comments

Wren: A classy little scripting language

https://wren.io/
127•Lyngbakr•4d ago•37 comments

Asbestosis

https://diamondgeezer.blogspot.com/2025/10/asbestosis.html
226•zeristor•18h ago•162 comments

Making the Electron Microscope

https://www.asimov.press/p/electron-microscope
62•mailyk•10h ago•8 comments

Feed the bots

https://maurycyz.com/misc/the_cost_of_trash/
158•chmaynard•14h ago•117 comments

Termite farmers fine-tune their weed control

https://arstechnica.com/science/2025/10/termite-farmers-fine-tune-their-weed-control/
3•PaulHoule•5d ago•0 comments

Eavesdropping on Internal Networks via Unencrypted Satellites

https://satcom.sysnet.ucsd.edu/
179•Bogdanp•6d ago•29 comments

Alzheimer's disrupts circadian rhythms of plaque-clearing brain cells

https://medicine.washu.edu/news/alzheimers-disrupts-circadian-rhythms-of-plaque-clearing-brain-ce...
162•gmays•9h ago•29 comments

Pico-Banana-400k

https://github.com/apple/pico-banana-400k
364•dvrp•1d ago•60 comments

Downloadable movie posters from the 40s, 50s, 60s, and 70s

https://hrc.contentdm.oclc.org/digital/collection/p15878coll84/search
414•bookofjoe•1w ago•80 comments

AI Mafia Network – An interactive visualization

https://dipakwani.com/ai-mafia/
82•dipakwani•3h ago•6 comments

Resource use matters, but material footprints are a poor way to measure it

https://ourworldindata.org/material-footprint-limitations
14•surprisetalk•12h ago•2 comments

Poison, Poison Everywhere

https://loeber.substack.com/p/29-poison-poison-everywhere
108•dividendpayee•4h ago•55 comments

Formal Reasoning [pdf]

https://cs.ru.nl/~freek/courses/fr-2025/public/fr.pdf
124•Thom2503•14h ago•27 comments

Milei Wins Mandate for Free-Market Revolution in Argentina's Election

https://www.wsj.com/world/americas/milei-wins-mandate-for-free-market-revolution-in-argentinas-el...
21•JumpCrisscross•1h ago•7 comments

Validating your ideas on strangers (2017)

https://jeremyaboyd.com/post/validating-your-ideas-on-strangers
61•tacon•2d ago•32 comments
Open in hackernews

Microsoft 365 Copilot – Arbitrary Data Exfiltration via Mermaid Diagrams

https://www.adamlogue.com/microsoft-365-copilot-arbitrary-data-exfiltration-via-mermaid-diagrams-fixed/
111•gnabgib•3h ago
https://web.archive.org/web/20251023095538/https://www.adaml...

Comments

simonw•2h ago
That site just gave me a 503 but here's the Internet Archive copy: https://web.archive.org/web/20251023095538/https://www.adaml...

This isn't the first Mermaid prompt injection exfiltration we've seen - here's one from August that was reported by Johann Rehberger against Cursor (and fixed by them): https://embracethered.com/blog/posts/2025/cursor-data-exfilt...

That's mentioned in the linked post. Looks like that attack was different - Cursor's Mermaid implementation could render external images, but Copilot's doesn't let you do that so you need to trick users with a fake Login button that activates a hyperlink instead.

luke-stanley•2h ago
The Lethal Trifecta strikes again! Mermaid seems like a bit of a side issue, presumably there are lots of ways data might leak out. It could have just been a normal link. They should probably look further into the underlying issue: unrelated instruction following.

Thanks for the archive link and the very useful term BTW! I also got 503 when trying to visit.

simonw•2h ago
I think they're doing this the right way. You can't fix unrelated instruction following with current generation LLMs, so given that the only leg you can remove from the trifecta is mechanisms for exfiltrating the data.

The first AI lab to solve unrelated instruction following is going to have SUCH a huge impact.

hshdhdhehd•1h ago
Not even humans can do it perfectly (hence social engineering)
binarymax•2h ago
> MSRC bounty team determined that M365 Copilot was out-of-scope for bounty and therefore not eligible for a reward.

What a shame. There’s probably LOTS of vulns in copilot. This just discourages researchers and responsible disclosure, likely leaving copilot very insecure in the long run.

CaptainOfCoit•2h ago
> There’s probably LOTS of vulns in copilot

Probably exactly why they "determined" it to be out of scope :)

candiddevmike•1h ago
It's irresponsible for any company to be using copilot with MS having this bug bounty attitude, IMO. Would be curious what other products are out of bounds so I know not to use them...
p_ing•1h ago
QQ for the LLM folks -- is this possibly due to the lack of determinization of LLM output?

If I code a var blah = 5*5; I know the answer is always 35. But if I ask an LLM, it seems like the answer could be anything from correct to any incorrect number one could dream up.

We saw this at work with the seahorse emoji question. A variety of [slight] different answers.

nawgz•1h ago
> If I code a var blah = 5*5; I know the answer is always 35

I greatly enjoy the irony here.

anonymars•1h ago
It's okay, we've replaced the Turing test with the em dash test
DrewADesign•48m ago
The em dash thing seems weird to me. The writing style guide for the college I attended as a freshman was big on them, and I never shook the habit. Not being able to easily conjure one was one of the biggest annoyances when I was forced to switch from macOS to windows.
dpark•41m ago
I find the em dash thing weird as well. I bunch of people who didn’t know what an em dash was a couple of years ago decided that it’s a signature LLM move.
tatersolid•12m ago
One of my first jobs was as the programmer/IT/graphics guy at a newspaper. Everybody there learned to use em-dashes properly and regularly, and followed other esoteric rules from the AP Stylebook that also appear in LLM output.

Which points to just how much unlicensed copyrighted material is in LLM training sets (whether fair use or not).

a-dub•2h ago
" ... BUT most importantly, ... "

i love the use of all capitals for emphasis for important instructions in the malicious prompt. it's almost like an enthusiastic leader of a criminal gang explaining the plot in a dingey diner the night before as the rain pours outside.

Nextgrid•1h ago
It’s both interesting to see all the creative ways people find to exploit LLM-based systems, but also disappointing that to this day designers of these systems don’t want to accept that LLMs are inherently vulnerable to prompt injection and short of significant breakthroughs in AI interpretability will remain hopelessly broken regardless of ad-hoc “mitigations” they implement.
narrator•1h ago
Prompt Injection is an interesting difference between human consciousness and machine "consciousness", or what people try and liken to it. A human can easily tell when information is coming from his memory or internal thoughts and when it is coming from a possibly less reliable outside source. Gaslighting is essentially an attempted prompt injection and is considered psychological abuse. Interestingly, people complain about AI gaslighting them and AI doesn't seem to think that's a problem.
lazyasciiart•49m ago
Isn’t that what marketing is?