OffByQuant

Newly discovered: established #AiSOC uses local AI for alert triage. v21 adds model provenance and closure accounting. Effectiveness unverified; review policy for automated closures. github.com/beenuar/aisoc #SOC #AISecurity

Geiger inventories local #AI agents and #MCP configuration, with baseline comparisons for drift. No model or #API account needed. It reads configuration, not runtime behavior. github.com/Atomburstofficial… #AISecurity #MCPSecurity

Review #AI-agent skills before installing them. SkillSpector checks for dangerous code, prompt-injection patterns and #MCP risks. Static scans need no paid model #API; findings still need review. github.com/NVIDIA/SkillSpect… #AISecurity #OpenSource

#Metis supports AI-assisted security code review and SARIF triage, with a documented local-model path through #Ollama. Validate findings yourself. github.com/arm/metis #AI #AppSec

Same Model, Different Engine: #Strata and #Maya Against Plain #llamacpp on One #NVIDIA #RTX5090 uncategorized.blog/posts/sam…

A Fork That Decodes Faster: #Basalt Against #Strata on One #RTX5090 uncategorized.blog/posts/a-f… #qwen #localLLM #llm #nvidia #inference

Crouching VRAM, Sleeping Dragon: How the Strata Family Makes Massive AI Models Run by Waking Only What They Need uncategorized.blog/posts/cro… #strata #glm #localLLM #inference #nvidia #rtx5090

Everyone budgets a quality tax for 4-bit quants. I went looking for it on #Qwen 3.8 27B: GPQA, IFBench, 95 real CVEs, RULER at 262K, KL-divergence. The tax never showed up — Q4 matches @Alibaba_Qwen 's published BF16 scores on a single #RTX5090. uncategorized.blog/posts/the… #localLLM