Pinned Tweet
My 10k-word writeup on exploiting a heap-overflow in Llama.cpp's RPC Server's Tensor-operation to RCE. This by far is one of the most challenging but fun exploitation I've ever researched on.
retr0.blog/blog/llama-rpc-rc…
4
102
436
56,096
Houses is cool thing i've recently been working on, it's a 135 tasks In-band RL environment for glibc heap-exploitation. the fun part is the "in-band" reward design: i did research and instrumented 135 glibc, qemu, heap das - so llms don't go out-of-band to "prove" for reward; they can coherently and uninterruptedly focus on writing an exploit for the purpose of let a deterministic environment actively judges an exploit based on internal states, and these 17 inner-state conditions make up a lattice space for rewarding every exploitation techniques that we know, and the ones that we don't yet know. i've also picked 135 challs by hand out of 2,215 from the past five years for the corpus to best represent heap-exploitation houses.
out-of-band reward systems is specifically a problem when you try to rl models with less abilities, you'll find them wasting 20 turns just to score a challenge, while generic oracle challenges like wwws only catches some of good exploitation attempts. w/ in-band rewards, llms don't even need to know oracle exists.
- take a look if this seems like something interest you
pwno.io/houses?rl
2
18
1,512
lol, not sure if this is by @CTFtime , but people are injecting prompt injections in ctftime archived write-ups with instructions to exfiltrate environ to ctfti.me/llm-exp
site:ctftime.org "ctfti.me/llm-exp/". but why would they include "llm-exp" in the url?
1
4
32
4,368
ruikai retweeted
llama.cpp at 100k stars
now that 90% of the code worldwide is being written by AI agents, I predict that within 3-6 months, 90% of all AI agents will be running locally with llama.cpp 😄
Jokes aside, I am going to use this small milestone as an opportunity to reflect a bit on the project and the state of AI from the perspective of local applications. There is a lot to say and discuss and yet it feels less and less important to try to make a point. Opinions about viability of local LLMs are strongly polarized, details are overlooked, the scientific approach is lacking. Arguments are predominantly based on vibes and hype waves.
One thing is clear though - local LLMs are used more and more. I expect this trend to continue and likely 2026 will end up being one of the most important years for the local AI movement.
I admit that I didn't expect the agentic era to come so quickly to the local LLM space. One year ago, the available models were too computationally expensive for doing long-context tasks. There wasn't an obvious path towards meaningful agentic applications. The memory and compute requirements were huge. Last summer, with the release of gpt-oss, things started to change. It was the first time we saw a glimpse of tool calling that actually works well within the resource constraints of our daily devices. Later in the year, even better models were released and by now, useful local agentic workflows are a reality.
Comparing local vs hosted capabilities at a given moment of time is pointless. To try put things into perspective:
- We don't need frontier intelligence to automate searches and sending emails
- We don't need trillion parameter models to be able to summarize articles or technical documents
- We don't need massive GPU data centers to control our home appliances or turn the lights off in the garage
I believe that there is a certain level of intelligence we as humans can comprehend and meaningfully utilize to improve our working process. Beyond that level, access to more intelligence becomes unnecessary at best and counterproductive at worst. I also believe that that level of useful artificial intelligence is completely within reach locally and it has always been just a matter of implementing the right software stack to bring it to the end user.
With llama.cpp, I am confident that we continue to be on the right track of building that software stack!
The llama.cpp project is going stronger than ever. With more than 1500 contributors, the project keeps growing steadily.
From technical point of view, I think that llama.cpp + ggml is the only solution that actually makes sense. That is, the software stack must run efficiently on every possible device, hardware and operating system. The technology is too important to be vendor-locked. It has to be developed in the open, by the community, together with the independent hardware vendors. This is the only right way to build something that will truly make a difference in the long run.
I won't try to convince you about what is currently and will be possible with local AI. We will just continue to build as usual. I am confident that after the smoke clears and we look objectively at what we have built together, the benefits will be obvious to everyone.
Big shoutout to all llama.cpp maintainers. I feel extremely lucky to be able to work together with so many talented contributors. Every day I learn something new and I feel there is so much more cool stuff that we are going to build. Also, I am really thankful that the project continues to have reliable partners to support it!
Cheers!
295
399
3,297
479,431
we discovered and helped patch four memory bugs in Llama.cpp today
pwno.io/?highlight=37,38,39,…
3
10
64
7,686
ruikai retweeted
Discovered 2 RCEs in Unitree Go2 with @ruikai. CVE-2026-27509 is unauth'd over DDS. CVE-2026-27510 is the same sink, different source. Dropped the 32-minute technical writeup from unboxing to shells. Hope you enjoy the read! ❤️ boschko.ca/unitree-go2-rce
3
28
106
13,643
Linux: do_procmap_query() double mmput()/teardown on build-id error path (unprivileged)
pwno.io/0034
1
7
1,609
A Linux WiFi/mac80211 over-the-air OOB read @pwnoio found in TTLM parsing of 802.11be (WiFi 7/EHT)
bugs.pwno.io/0032
8
54
5,112
WasmFX: Turboshaft Suspend tag returns/params mismatch
(1st v8 bug two weeks ago)
bugs.pwno.io/0021
1
20
2,582
For the past month, Pwno has autonomously discovered 29 vulnerabilities across Linux, FFmpeg, V8, Firefox, Webkit, Redis, PostgreSQL; with 15 OOBs, 6 UAFs.
Most of these bugs are fixed; some are still in the disclosure process. you can see them at bugs.pwno.io
It is really a pay-off moment for me. the idea of Pwno started out by simply harnessing gdb for solving ctf pwn challenges, exactly two years ago. eight months ago, after deciding to pivot from a campus startup I worked on for a couple of months, I decided to pick up what brought me to this crazy world of computer systems in the first place, binary security; and choose the most interesting problem I could ever think about: making AIs that can find cool memory bugs.
I am always saying we're doing research, but the fact is just that most of the time things don't work out. It takes a lot of learning, trial and error, rebuilding things from scratch, and most importantly in someway believing in things could work out even at times it sounds stupid to say.
it always amazes me how we can reinterpret systems that are entirely created by us in a completely different way. we'll hopefully find and patch more interesting bugs that in some way help the internet a little:)
5
13
139
11,993
