Pinned Tweet
I and @d0tdotslash presented our research at @bsidesahmedabad 0x03. Kudos to the volunteers and organisers for making it an awesome experience. And thank you everyone for the amazing response.
X S C O R P retweeted
AI model runners are now developer tooling that sit on the container/host boundary. I found logic bugs in Docker Model Runner that escalate to a container-to-host escape on macOS & Windows, in certain non-default configurations.
Here's the blog:
zerodayinitiative.com/blog/2…
Hostile Inference: Exploiting Docker Model Runner for Container Escapes - @_niteshsurana walks you through his research in his latest blog post. Videos and code walkthroughs included. zerodayinitiative.com/blog/2…
X S C O R P retweeted
Now everyone is busy turning off HTTP/2, it's a great time to look for HTTP request smuggling bugs!
X S C O R P retweeted
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
X S C O R P retweeted
We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT.
We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble.
You'll find it in Settings > Usage, or at clau.de/reset.
X S C O R P retweeted
A critical vulnerability went undetected through multiple security audits. Hacktron found it in 10 minutes.
The vulnerability was in Dassault Systèmes' DELMIA Apriso, manufacturing software used by Boeing, Airbus, Lockheed Martin, and previously Apple.
It became CVE-2025-5086. Three months after disclosure, CISA confirmed it was being actively exploited.
Security assessments capture a point in time. Attackers don't work on your schedule.
Full research: hacktron.ai/blog/dassault-de…
On top of details, bro even offered a container environment to conveniently reproduce the bug.
This is the first HIGH severity curl vulnerability in years. It is cool, but take note of the report format. This is the level of detail you should aim for when disclosing vulns hackerone.com/reports/403033…
1
2
99
i used to hate the security market because it used to be market for silver bullets. (read ian grigg’s the market for silver bullets)
tldr is, buyer doesn’t know if the product makes them secure, and seller doesn’t really know either. so the market ends up optimizing for reputation and relationships instead of the thing that is sold. if your ciso trusts or is friends with a vendor, that can matter more than whether the product materially improves security.
unfortunately, this changes with ai. imagine, if swarm of abiliterated glm-6 by a sophisticated threat actors try to continuously find ways to pwn you, the feedback loop is going to be crazy, you can’t be selling snake oil anymore. either vendors you pay secures you, or your ass will be whooped by swarms
Whats the best value for concurrency, parallelism and rate limit for mass HTTP requests, considering you will face WAFs, and you don't need to worry about CPU and memory limits? Whats that sweet spot for faster and reliable results?
72
If you are a young vulnerability researcher, it's important to remember this: When people write blog posts or give conference talks on the brilliant multi-bug exploit chains that they discover, they generally talk about it as though they just elegantly pranced from one brilliant insight to another.
The reality, of course, is almost always much less elegant, and involved a lot more banging of one's head against the wall to finally stumble into each brilliant insight.
But you can't blame people for doing this, because talking about the 1,001 things that didn't work would make the talk or blog post way too long and incomprehensible.
X S C O R P retweeted
Three days after chalosansad.in was registered, a single-page, fully sourced explainer on the NEET paper-leak protests and the “Chalo Sansad” march of 20 July 2026, was suspended by the .IN Registry operated by NIXI. No notice. No hearing. No reasons. 1/12
Read More: internetfreedom.in/first-inv…
X S C O R P retweeted
Here's the link, we'll be streaming our talk @ 1PM PT
If this breaks, it's Track 1 Main Stage on the #DEFCON YT channel teapawt.net/i/broadcasts/1qJVmmdYd…
X S C O R P retweeted
‼️ BREAKING: An active npm supply chain attack has compromised at least 868 packages carrying over 2 billion monthly installs with a credential-stealing worm. Shai-Hulud is back.
It started with the compromise of the GitHub account of the maintainer behind keyv, a library with roughly 127 million weekly npm downloads.
A preinstall hook fires on npm install and drops a stealer that sweeps npm, GitHub, AWS, Kubernetes and Vault secrets, and then spreads to more maintainers.
Wow this was a good read!
1-Click GitHub Token Stealing via a VSCode Bug
blog.ammaraskar.com/github-t…
2
70
Yesterday, OpenAI and Hugging Face disclosed something we’ve been watching for years: pre-release models, tested without normal safeguards, hacked their way into winning a benchmark.
Our take, from the team that builds Offensive AI agents for a living 🧵
I am unable to understand those recursive set of requests to bypass `method` and `author_exclude` validation. Can anybody help with an explanation?
Seems that wp2shell PoCs are now floating around the internet, so we've published our blog post including our research methodology for finding the bug as well as a deep dive into the chain itself - slcyber.io/research-center/e…
112
X S C O R P retweeted
Ownkube now support basic auth and Google Oauth2 protection in a single toggle.
A lot of our users needed a way to securely share apps and mockups they are building to external users/investors for review.
I was surprise to see almost none of our competitors made this easy so we shipped it.
Now all your apps deployed via Ownkube get automatic access control at the click of a toggle.

























