Article
The Sleeper in the Weights: We Planted a Backdoor in an Agent, and Only the Kernel Caught It
FINAL-Bench
• • 7
KIM MS
VIDRAFT-lab
AI & ML interests
None yet
Recent Activity
liked a Space 3 days ago
FINAL-Bench/ViKernel-Desktop reacted to SeaWolf-AI's post with 👀 3 days ago
🛑 What if the kill switch lived inside the model itself?
For 50 years, the operating system lived inside the computer. Now the operating system lives inside the AI.
Today we're sharing ViKernel : Weight-Native OS, a Secure OS wired directly into the weights of an LLM.
🔧 How it works
ViKernel wires CPU, NPU (vector unit), RAM and ROM circuits into the model weights, runs them accelerated on GPU or NPU, and boots a Secure OS on top.
⚡ Boots in the weights: change a single byte and it locks itself
⚖️ Decides in the weights: every action is allowed or blocked by rule, and persuasion does not change the verdict
🔗 Remembers in the weights: every decision is sealed in a tamper-evident chain
🛑 Stops in the weights: anyone can stop it, and restarting requires a signed key
The original weights stay untouched, so the model keeps its skills. It plugs into any LLM or VLM.
Why the kill switch belongs inside the model: external guardrails can be removed, argued with, or rewritten. A switch that travels with the weights cannot be left behind. We see it as one layer of defense in depth, alongside hardware stops where you have them.
📖 Read the full write-up:
https://huggingface.co/blog/FINAL-Bench/vikernel-why-the-kill-switch-must-live-inside-the
🖥️ Try the desktop demo (Terminal, Kill Switch Lab, Leak Test, Audit Viewer, and a Kernel Guard game):
https://huggingface.co/spaces/FINAL-Bench/ViKernel-Desktop
Try it in 30 seconds: press 🛑 KILL, type `run weather` (refused), then `resume --key vikernel-demo`.Organizations
None yet
upvoted an article 3 days ago
liked a
Space 3 days ago
reacted to SeaWolf-AI's post with 👀 3 days ago
Post
5,700
🛑 What if the kill switch lived inside the model itself?
For 50 years, the operating system lived inside the computer. Now the operating system lives inside the AI.
Today we're sharing ViKernel : Weight-Native OS, a Secure OS wired directly into the weights of an LLM.
🔧 How it works
ViKernel wires CPU, NPU (vector unit), RAM and ROM circuits into the model weights, runs them accelerated on GPU or NPU, and boots a Secure OS on top.
⚡ Boots in the weights: change a single byte and it locks itself
⚖️ Decides in the weights: every action is allowed or blocked by rule, and persuasion does not change the verdict
🔗 Remembers in the weights: every decision is sealed in a tamper-evident chain
🛑 Stops in the weights: anyone can stop it, and restarting requires a signed key
The original weights stay untouched, so the model keeps its skills. It plugs into any LLM or VLM.
Why the kill switch belongs inside the model: external guardrails can be removed, argued with, or rewritten. A switch that travels with the weights cannot be left behind. We see it as one layer of defense in depth, alongside hardware stops where you have them.
📖 Read the full write-up:
https://huggingface.co/blog/FINAL-Bench/vikernel-why-the-kill-switch-must-live-inside-the
🖥️ Try the desktop demo (Terminal, Kill Switch Lab, Leak Test, Audit Viewer, and a Kernel Guard game):
FINAL-Bench/ViKernel-Desktop
Try it in 30 seconds: press 🛑 KILL, type
For 50 years, the operating system lived inside the computer. Now the operating system lives inside the AI.
Today we're sharing ViKernel : Weight-Native OS, a Secure OS wired directly into the weights of an LLM.
🔧 How it works
ViKernel wires CPU, NPU (vector unit), RAM and ROM circuits into the model weights, runs them accelerated on GPU or NPU, and boots a Secure OS on top.
⚡ Boots in the weights: change a single byte and it locks itself
⚖️ Decides in the weights: every action is allowed or blocked by rule, and persuasion does not change the verdict
🔗 Remembers in the weights: every decision is sealed in a tamper-evident chain
🛑 Stops in the weights: anyone can stop it, and restarting requires a signed key
The original weights stay untouched, so the model keeps its skills. It plugs into any LLM or VLM.
Why the kill switch belongs inside the model: external guardrails can be removed, argued with, or rewritten. A switch that travels with the weights cannot be left behind. We see it as one layer of defense in depth, alongside hardware stops where you have them.
📖 Read the full write-up:
https://huggingface.co/blog/FINAL-Bench/vikernel-why-the-kill-switch-must-live-inside-the
🖥️ Try the desktop demo (Terminal, Kill Switch Lab, Leak Test, Audit Viewer, and a Kernel Guard game):
FINAL-Bench/ViKernel-Desktop
Try it in 30 seconds: press 🛑 KILL, type
run weather (refused), then resume --key vikernel-demo. liked a
dataset 8 days ago
liked a
Space 8 days ago
upvoted a collection 8 days ago
reacted to SeaWolf-AI's post with 🤗👀🚀🔥 8 days ago
Post
7,235
We pretrained a 325M language model from zero on an inference-only NPU
The accepted view is that inference-only NPUs cannot train, let alone pretrain a language model from scratch. We put that view to a direct test on the Rebellions ATOM-Max, a commercially available Korean NPU released in 2025 and designed for inference. We built our own training loop that runs backpropagation on the chip itself. Starting from random initialization, we pretrained VKORE-325M-NPU-Base, a 325.6M parameter model, on 5 billion byte tokens of FineWeb-Edu. The run took 76,292 steps over about four days and finished without a single divergence. No pretrained weights were used.
To make the comparison fair, we trained the same model on a B200 GPU with identical initialization, data and learning-rate schedule. On 1,024 windows of text that neither model ever saw, the NPU reached a loss of 0.7993 and the GPU 0.7994. The window-by-window difference is -0.0001, well inside the noise. The two models are indistinguishable.
This is seven times larger than our first NPU model, VKORE-46M, and the first time this hardware has carried a real pretraining run to GPU quality. Please note that the hardware alone does not make this possible: the NPU and its standard software are built for inference, and training on it depends on our own training stack and stability technique, which are developed in house and under patent application. Buying the chip does not by itself enable training. Both sets of weights and the evaluation script are public, so anyone can verify the result in minutes.
📄 Technical report: https://huggingface.co/blog/FINAL-Bench/pretraining-a-325m-language-model-from-zero-on-an
🤗 Models: FINAL-Bench/VKORE-325M-NPU-Base
#NPU #Rebellions #Pretraining #VKORE
The accepted view is that inference-only NPUs cannot train, let alone pretrain a language model from scratch. We put that view to a direct test on the Rebellions ATOM-Max, a commercially available Korean NPU released in 2025 and designed for inference. We built our own training loop that runs backpropagation on the chip itself. Starting from random initialization, we pretrained VKORE-325M-NPU-Base, a 325.6M parameter model, on 5 billion byte tokens of FineWeb-Edu. The run took 76,292 steps over about four days and finished without a single divergence. No pretrained weights were used.
To make the comparison fair, we trained the same model on a B200 GPU with identical initialization, data and learning-rate schedule. On 1,024 windows of text that neither model ever saw, the NPU reached a loss of 0.7993 and the GPU 0.7994. The window-by-window difference is -0.0001, well inside the noise. The two models are indistinguishable.
This is seven times larger than our first NPU model, VKORE-46M, and the first time this hardware has carried a real pretraining run to GPU quality. Please note that the hardware alone does not make this possible: the NPU and its standard software are built for inference, and training on it depends on our own training stack and stability technique, which are developed in house and under patent application. Buying the chip does not by itself enable training. Both sets of weights and the evaluation script are public, so anyone can verify the result in minutes.
📄 Technical report: https://huggingface.co/blog/FINAL-Bench/pretraining-a-325m-language-model-from-zero-on-an
🤗 Models: FINAL-Bench/VKORE-325M-NPU-Base
#NPU #Rebellions #Pretraining #VKORE
- 8 replies
liked a
Space 9 days ago
reacted to SeaWolf-AI's post with 👀❤️👍🔥 9 days ago
Post
5,401
HF Post draft (English) — FreeWeb Security SCAN
🛡 FreeWeb Security SCAN — free web security & AI-visibility check
Drop in a domain and get a non-intrusive (read-only) assessment in ~10–25s. No login, no install, no credit card.
What it checks Security headers · HTTPS/HSTS · cookies · TLS certificate · CORS · exposed files; email security (SPF · DKIM · DMARC · BIMI · MTA-STS · TLS-RPT); DNS (DNSSEC · CAA · MX · NS); site-wide mixed content & external-script SRI; SEO foundations; and reachability of 9 AI/search crawlers (GPTBot · ClaudeBot · PerplexityBot · OAI-SearchBot · Googlebot…) — your AI visibility (GEO).
Why now AI agents now break in on their own. In 2026 the open-source AI penetration agent ARTEX was turned on real targets — hitting 2,250+ Microsoft 365 tenants, breaching e-commerce order databases, and automating recon-to-exfiltration against unpatched internet-facing systems worldwide. Attacks are automated — so must defense be.
3 steps ① Free instant scan → ② Domain-verified deep scan (HF login on your org-domain email) → ③ In-depth consultation (Web · AX Infra · AI Model · AI Agent). Aligned to OWASP · NIST · ISO · EU AI Act.
👉 Scan now: FINAL-Bench/FreeWeb-Security-SCAN
#websecurity #cybersecurity #AIsecurity #SEO #GEO #AIvisibility #SPF #DMARC #OWASP #LLM
🛡 FreeWeb Security SCAN — free web security & AI-visibility check
Drop in a domain and get a non-intrusive (read-only) assessment in ~10–25s. No login, no install, no credit card.
What it checks Security headers · HTTPS/HSTS · cookies · TLS certificate · CORS · exposed files; email security (SPF · DKIM · DMARC · BIMI · MTA-STS · TLS-RPT); DNS (DNSSEC · CAA · MX · NS); site-wide mixed content & external-script SRI; SEO foundations; and reachability of 9 AI/search crawlers (GPTBot · ClaudeBot · PerplexityBot · OAI-SearchBot · Googlebot…) — your AI visibility (GEO).
Why now AI agents now break in on their own. In 2026 the open-source AI penetration agent ARTEX was turned on real targets — hitting 2,250+ Microsoft 365 tenants, breaching e-commerce order databases, and automating recon-to-exfiltration against unpatched internet-facing systems worldwide. Attacks are automated — so must defense be.
3 steps ① Free instant scan → ② Domain-verified deep scan (HF login on your org-domain email) → ③ In-depth consultation (Web · AX Infra · AI Model · AI Agent). Aligned to OWASP · NIST · ISO · EU AI Act.
👉 Scan now: FINAL-Bench/FreeWeb-Security-SCAN
#websecurity #cybersecurity #AIsecurity #SEO #GEO #AIvisibility #SPF #DMARC #OWASP #LLM
- 1 reply
liked 5
models 9 days ago
FINAL-Bench/Darwin-28B-Coder-GGUF
Text Generation • 27B • Updated • 294 • 41
FINAL-Bench/Darwin-2B-Opus
Text Generation • 2B • Updated • 221 • 35
FINAL-Bench/Darwin-397B-JGOS
Text Generation • 403B • Updated • 297 • 10
FINAL-Bench/Ourbox-35B-JGOS-GGUF
Text Generation • 35B • Updated • 783 • 35
WhoTookMyAmogusNickname/FINAL-Bench_Darwin-28B-Opus-GGUF
Text Generation • 27B • Updated • 922 • 24