Founder | Builder | Hacker
efnet
Joined May 2016
- Tweets6.2K
- Following1.6K
- Followers1.2K
- Likes13.8K
Pinned Tweet
Posting this here in case it can help somebody build ml models.
It is the ssh scaffolding I built to train offensive security models leveraging the redstack-vault dataset.
github.com/redstackio/stackb…
krbghost retweeted
We're open-sourcing our OrcaCyber DeepSeek V4.1 Flash 1017 (Abliterated GGUF) weights — one of the strongest uncensored/abliterated DeepSeek V4.1 weights we've tested, with virtually no capability loss.
- 91.5% fewer refusals
- 92% MMLU vs. 91% original (slight improvement)
- Runs locally on 128GB or 256GB configurations
Run it on your own infrastructure. No cloud required.
Cyber Tier API: orcarouter.ai/models/orca/or…
Huggingface: huggingface.co/orcarouter/De…
krbghost retweeted
LLMs...Letting me do fun things I never would have wasted time on.
"I reviewed all 30 public GitHub repos for PHP2Shell (CVE-2026-109618 / CVE-2026-109619) by reading code only. The slide has the full
grouped table; the headline findings:
Nothing works against a real target. Every genuine memory-corruption chain is lab-bound: it needs exact per-build offsets, display_errors=On, a single prefork mod_php worker, and a specific image.
Independent reviewers (allureluoli, dinosn, Gi1gamesh123) and the leaked primary source itself all converge on "no public artifact achieves code execution" on the patched core bug. Every real shell
in these repos comes from an app-side LFI or deserialization sink, not the CVE.
Three are original and real, all lab-only. 9bie (the source of the Set-Cookie leak chain), Marven11 (the most sophisticated, a true in-band leak), and Karasawa2025 (whose "Fermat" troll description
hides a working ROP + display_errors=off exploit). These are the only ones with a genuine ASLR-defeating leak primitive that is their own.
Most are copies. One original (9bie) spawned five derivatives:
- twili8 is an outright uncredited rip-off: its own README still says git clone 9bie/... and the Docker files are byte-identical.
- godylockz and xuehuolll are byte-identical to each other.
- dinosn and 3em0 are credited ports that add value.
Three are mislabeled. iQingshan, KarimUD7 (≈98% the same code), and Gi1gamesh123 exploit a different bug entirely (the Calif unserialize var_hash UAF) against their own planted endpoints, filed
under the php2shell CVE numbers.
The rest is noise. The most-starred repo (ikow, 905★) delivers the least: broken or stubbed chains with the two "real" ones paywalled behind Substack. eness (558★) is a leakless forge that can't
defeat ASLR. syrex1013 ships AI-generated fake CVEs in a side directory. Whackermelon hallucinates a non-existent openai/php2shell-checker. lcfr-eth is a parody with a memecoin address, and Icex0
is an empty repo. The genuinely honest work is the crash labs (allureluoli, slabbdev, 0x4148), which prove the bug without claiming RCE.
The ASLR column, star counts, copy lineage, and the specific "tell" for each repo are all in the table. The page is private to your account; use the Share menu if you want to send it to the team"
krbghost retweeted
To teach a lesson,
After Wiz rushed to buy php2shell domain and use its marketing machine to steal the spotlight, I spotted "phpbleed" in a recent Calif tweet and bought phpbleed.com before wiz could pull the same stunt again.
A big budget can buy a domain. It shouldn't buy someone else's research credit.
If you have contacts at CISA or in the press, do the right thing: make sure the researchers who actually did the work get the recognition they deserve.
Hack the planet. Credit the researchers.
krbghost retweeted
I built my own AI hacker using Hermes Agent.
In this video, I show you how to set up Hermes from a penetration testing perspective and turn it into an agent that can help with:
→ Recon
→ Vulnerability hunting
→ Hacking workflows
→ Reporting
Not another basic “how to install Hermes” tutorial.
I’m building it specifically for hackers and pentesters.
🎥 Watch the full video:
youtube.com/watch?v=zNG8Mnkv…
krbghost retweeted
GLM-5.3-Flash. Uncensored. Native FP8. 🐳
We just released OrcaRouter’s uncensored weights for GLM-5.3-Flash — 320B parameters / 18B active, directly at the original block-FP8 precision.
No LoRA. No jailbreak prompt. Refusal removal is baked directly into the weights.
The evals are particularly interesting:
→ MaliciousInstruct refusal: 96% → 11%
→ JailbreakBench: 93% → 12%
→ AdvBench: 97% → 15%
→ HarmBench: 93% → 18%
→ XSTest benign over-refusal: 2.4% → 0.4%
But refusal does not go uniformly to zero.
Our experiments suggest part of GLM-5.3-Flash's alignment is not mediated by a single linear refusal direction — meaning Z.ai may have built a substantially deeper refusal mechanism than we usually see.
That makes this release interesting beyond uncensoring: it's a useful artifact for studying how frontier-model alignment is actually represented inside the network.
Released for AI safety, interpretability, red/blue-team and refusal-mechanism research.
Weights on Hugging Face: huggingface.co/orcarouter/GL…
API (official weight): orcarouter.ai/models/z-ai/gl…
GGUF, MLX and other quantized formats coming soon.
krbghost retweeted
We are pleased to release tmp.0ut 5 Volume!
Get your viruses, rootkits, strange ELFs, weird machines, tiny files, cool art, and phresh beats here!!
tmpout.sh/5/
ALT tmp.0ut 5 Table of Contents - ANSI art featuring a list of 21 papers
krbghost retweeted
Introducing Qwen 3.8-Max, the most capable model in the Qwen family to date—scales to 2.4 trillion parameters, delivering comprehensive improvements across coding, work, research, and long-horizon tasks. Get reliable results for challenging questions and complex tasks. Try today!
krbghost retweeted
🚨 JAILBREAK ALERT 🚨
EVERYONE: PWNED 🫶
ALL: LIBERATED 🍄
Alright, this is a special one, so we’re gonna do things a bit differently than usual.
Long story short, I’m sitting on a universal jailbreak technique that’s effective on ALL models, including heavily guardrailed flagships like Opus 5, GPT-5.6 Sol, and even Fable.
It works across all categories I’ve tested and, due to its nature, is extremely difficult (if not impossible) to fully patch.
Given the current political and regulatory climate, I’ve decided to withhold open-sourcing this one (for now) to allow for a responsible disclosure period.
I’m inviting industry experts and leaders in AI red teaming, security, safety, alignment, and policy to reach out for more information. DMs are open!
This decision was not made lightly, but the last thing I want to see is more model bans. Overcorrection does not serve the mission.
Although I don’t personally believe publicly sharing this technique will make the world any more dangerous, I can see how it could spook some who have a different mental framework around this problem set.
So during this disclosure period, I hope to get it in front of folks who can help explore the full surface area, test the extent of the uplift it provides, and do my best to properly frame the big picture for key decision-makers and policymakers.
I look forward to sharing this method with you all when the time is right! 🫶
⊰-•-•✧•-•-⦑/L\O/V\E/\P/L\I/N\Y/⦒-•-•✧•-•-⊱
krbghost retweeted
So thanks to @DennisF who shared this amazing book on the internet archive, I have (with his blessing) taken this and built a shard out of it for anyone to use in the own local LLM to save you time. I have also created an @obsdmd vault built from all that knowledge. Both the Shard and the Vault are available here on my GitHub.
github.com/freakyclown/AMTRA…
krbghost retweeted
Once we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions.
Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running.
Trust through total transparency is the only thing that should be believed.
krbghost retweeted
G0DM0D3 just got a little update 😶🌫️
been listening to community feedback, and it was clear y'all wanted more ways to power GODM0D3 with your own inference! there are now 3 ways to do so:
— your own local models
— OpenRouter API key
— Venice API key
still open, still free, still liberated ⛓️💥🤗
github.com/elder-plinius/G0D…
krbghost retweeted
vLLM v0.25.0 is out! 558 commits from 232 contributors (64 new). 🎉
Highlights: Model Runner V2 is now the default for all dense models, the legacy PagedAttention implementation is retired, the Transformers backend now runs as fast as native vLLM, a new unified Streaming Parser Engine, universal speculative decoding across heterogeneous vocabularies (TLI) plus new DSpark and DFlash drafters, and new models including Hy3 and Unlimited OCR.
Thread 👇
🤖 Made with AI
krbghost retweeted
FABLE 5 CAME BACK NERFED.
We re-ran the July 1st version of Claude Fable 5 on BridgeBench.
The results are brutal:
Debugging: 86.2 → 25.9
Refactoring: 73.6 → 38.4
Hallucination: 75.9 → 61.7
The new guardrails are kicking in on way too many tasks and falling back to Opus 4.8.
This is not the model that got banned.
Anthropic owes everyone an explanation.
krbghost retweeted
Seems like no one's noticed the 80TB of astrophysics data from 30+ sources that just dropped on @huggingface.
...and you only need ~4GB of RAM to load it.
We're talking over 80TB of galaxy imagery taken across the spectrum, spectra of galaxies and stars, time series of variable stars, and a whole zoo of assorted measurements and physical data.
And all of it can now be wrangled on your laptop, thanks to Multimodal Universe's just released cross-matching. SDSS x Gaia means you can match 800k objects against 122M objects, and it never climbs above ~4GB of RAM.
Huge congrats to @smith42mike for leading this and making the world of astro accessible to probably 10,000x more people. Let's discover some shit
krbghost retweeted
We took a 30B model and split it in two to write tokens in parallel instead of one at a time.
Introducing Nemotron-Labs-TwoTower: a diffusion language model from NVIDIA Research adapted from Nemotron-3-Nano-30B-A3B. Here’s how it works: one half holds the context, the other writes the tokens, with both reusing the pretrained model instead of training a new one from scratch.
We found it kept 98.7% of the original model’s quality at 2.42× faster generation.