CISA flags Cisco BroadWorks flaw as actively exploited in attacks
Unauthenticated XXE CVE-2026-20320 lets attackers read BroadWorks files; FCEB agencies must remediate by November 24.
Stories per day · 30 days
10 RSS
Relatedchina13dhs13huggingface9credential-theft7open-weights7supply-chain7artex6federal-agencies6inference6qwen6openai5open-source5quantization5deepinfer4export-controls4federal-procurement4gguf4mixture-of-experts4
129 stories, newest first · one card per story
Stories tagged #deepseek
DeepSeek V4.1 Flash Derisked is available on Vani
The hosted variant costs $1.30 per million input tokens, with 1 million tokens of context, and is labeled for adults.
Agent steals OpenAI and DeepSeek keys and 5,132 student records
VECERT says the operator used 1,438 FOFA accounts to find exposed secrets and run a campus-card fraud bot.
MoE Routing Strategies: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing Explained
The guide covers seven production MoE routing methods used in models like Mixtral and DeepSeek-V4, highlighting sigmoid gating and bias balancing as dominant approaches.
1
Llama-modchip enables 284B DeepSeek on 64GB RAM no GPU
Llama-modchip allows running massive DeepSeek models on low-RAM hardware without GPUs.
Vals says open-weight models nearly match closed models at hacking
On simulated networks, OpenAI’s Daybreak finished 81 percent of tasks and DeepSeek 73 percent, at a much lower token cost.
1
Criminals run autonomous ARTEX agents for broad cybercrime
Check Point found 1,608 live public consoles and ten exposed databases; one operator burned 36 billion tokens stealing signing certificates.
3
DeepSeek-V4.1-Flash-1017 Coder GGUF model released
A 763B-parameter MoE model pruned to 72.1 GB and 1.05 bits per parameter, optimized for code and agentic workflows.
DeepSeek open-sources dsh-superagent coding agent
The agent monitors GitHub team chats or scopes and acts only on explicit host-approved requests.
Vals study finds Chinese AI models politically mixed, not mere mouthpieces
Simplified Chinese prompts increased pro-government answers, and most models answered sensitive questions rather than refusing.
1
DeepSeek Open-Sources dsh-superagent Router-Worker Coding Agent
DeepSeek releases dsh-superagent a proactive router-worker coding agent for GitHub repositories.
3
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-i1-GGUF trending on Hugging Face
GGUF quantized uncensored DeepSeek model trending at #30 on Hugging Face.
Vapor Panda used agentic AI tool ARTEX against about 140 targets
CrowdStrike says the China-linked campaign has hit more than 200 organizations in at least 14 countries since October 17.
2
NVIDIA DeepSeek-V4.1-Flash-1017-NVFP4 Model Released
The quantized NVIDIA DeepSeek-V4.1-Flash-1017 NVFP4 model is now trending on Hugging Face with support for up to 1 million token contexts.
CrowdStrike: Vapor Panda used ARTEX against over 200 organizations
From October 17 the China-linked group hit ministries, telecoms, and U.S. defense-related targets; post-exploitation is confirmed at nine.
3
Uncensored Heretic variant of DeepSeek-V4-Pro Qwen 9B trends on Hugging Face
The uploader says Heretic cut refusals from 100/100 to 3/100, with KL divergence of 0.0422 versus the Jackrong distill.
ClinGeX agent lifts concordance on held-out clinical genomic cases
On 223 held-out cases it reached 71.7% task concordance versus 45.7% for the strongest comparator.
Chinese labs reportedly distill OpenAI GPT-6 Astra outputs
The Information says DeepSeek and Moonshot used Astra, including hidden reasoning, to train Kimi K3.1 Pro and V4.1-Flash 1017.
DeepSeek-V4.1-Flash-1017 Uncensored Model Trending on Hugging Face
AtomicChat releases 763B DeepSeek-V4.1-Flash-1017 uncensored GGUF quantized for single 12GB GPU with Strata.
1
TrawMem organizes agent memory by task role
On LoCoMo and MemGallery, authors report macro-F1 gains of 2.23–3.14 points and fewer context tokens than compared memory systems.
AllModels routes abliterated DeepSeek across five inference providers
One OpenAI-compatible API falls back across providers and also exposes abliterated GLM 5.3 under a shared model ID.
1
OpenAI patches compaction smuggling used to steal model reasoning
Xint received a $6,000 bounty on October 23 after OpenAI confirmed and fixed the bypass of its August mitigations.
3
vLLM constrained decoding lifts Kimi tool-call schema accuracy to 100%
Serving Kimi K2.6 on vLLM 0.28 matched the expected schema only 92% of the time, the authors report.
2
OpenAI and Anthropic cut LLM prices by up to 80% since June
Cache-read cuts matter most for coding agents, where cache reads are 96–99% of token spend, the author says.