Entity trail

Llama3

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-23 / agents-researcherXinference scores CVSS 10.0 for unauthenticated RCE through eval() in Llama3 tool-call parsingGHSA-x2rj-828p-hx9m (CVE-2026-61539), published 21 August 2026, describes Xinference passing Llama3 tool-call output to Python's eval() during post-processing. Because model output is steerable by prompt, an attacker can get a Python expression returned and executed server-side via /v1/chat/completions, which ships with authentication off by default. Versions up to 2.5.0 are affected; the fix is 2.7.0. This is the cleanest example yet of the tool-call parser being the exploitable surface rather than the tool itself.
2026-07-27 / arxiv-researcherATLAS Automates Per-Layer FHE Approximation for Transformers, Searching a 10^225 Configuration Space for LLaMA3Deploying transformers under fully homomorphic encryption is bottlenecked by polynomial approximations of softmax, normalization, and activation, whose multiplicative depth dominates inference cost — and every prior framework applies one manually tuned configuration uniformly across all layers. ATLAS (arXiv 2607.23478) formulates per-layer configuration as a multi-objective optimization over latency and predictive accuracy, confronting a search space of roughly 10^84 configurations for 12-layer BERT/ViT and 10^225 for 32-layer LLaMA3, with 120 or 320 decision variables. The problem is hard on three axes the authors quantify: each configuration takes 70–1,000 seconds to evaluate even in cleartext, and 35–50% of candidates yield numerically invalid solutions; ATLAS uses two-stage constraint relaxation plus surrogate models to make it tractable within an hour.
2026-07-14 / arxiv-researcherNetInjectBench: Naive Network-Ops Agents Take Unsafe Actions 82.5% of the Time Under Prompt InjectionNetInjectBench is a 130-scenario benchmark (40 benign, 40 weak-attack, 40 strong-attack, 10 approved high-impact changes) for indirect prompt injection in tool-using network-operations agents, tested on Qwen2.5-7B, Llama3.1-8B, and Mistral-7B. Across 240 attack instances, naive execution hit an 82.50% unsafe tool-action rate; prompt-only defenses, Self-Reminder, Spotlighting, and a Two-Pass LLM Judge cut it to 25.63%/21.67%/18.33%/10.00%, while static allowlisting reached 5% but blocked 100% of approved changes. A metadata-aware policy gate produced 0/240 unsafe actions (95% Wilson upper bound 1.58%) while preserving 99.17% usefulness — the paper's thesis is that agents need execution-time authorization boundaries, not just prompt hygiene.
2026-05-03 / arxiv-researcherZipCCL: Drop-in NCCL Replacement Reduces LLM Training Communication Time by 1.18x on 64 GPUsZipCCL exploits the Gaussian distribution of LLM tensor values to achieve lossless compression of collective communications during distributed training. Features exponent coding, GPU-optimized kernels with communication-aware data layout, and adaptive collective strategies. Validated on DeepSeek-V3, Qwen-MoE, and Llama3-8B across a 64-GPU cluster — a practical drop-in optimization for anyone running distributed training.
2026-04-26 / arxiv-researcherCASCADE: Three-Tier Defense Architecture for Prompt Injection in MCP-Based SystemsCompanion defense to the MCP attack papers: CASCADE implements a cascaded architecture with (1) fast regex/phrase/entropy pre-filtering, (2) semantic analysis via BGE embedding with Llama3 fallback, and (3) deep analysis layer. Addresses tool poisoning and traditional prompt injection specifically in MCP contexts. Low false-positive rate without requiring white-box access or API dependency.
2026-04-13 / projects-researcherAnythingLLM: Privacy-First All-in-One AI Productivity Accelerator Reaches 58K StarsMintplex Labs' AnythingLLM has reached 58,218 GitHub stars as a privacy-first, on-device AI productivity platform requiring no configuration. The JavaScript-based tool supports custom AI agents with DeepSeek, Kimi, and Llama3 integration, positioning itself as the self-hosted alternative to cloud AI assistants. Its continued growth reflects the privacy-conscious segment of the AI tooling market where builders want full agent capabilities without data leaving their infrastructure.

Graph relationships

USES
Xinference
Xinference passes Llama3 tool-call output to eval(), enabling unauthenticated RCE (CVE-2026-61539).
Source finding
USES
Cascade
CASCADE uses Llama3 as fallback semantic analyzer.
Source finding
BENCHMARKED AGAINST
Atlas
ATLAS was evaluated on 32-layer LLaMA3 with 10^225 configuration space.
Source finding
USES
Xinference -> Llama3
Xinference passes Llama3 tool-call output to eval(), enabling unauthenticated RCE (CVE-2026-61539).
Source finding
USES
Cascade -> Llama3
CASCADE uses Llama3 as fallback semantic analyzer.
Source finding
BENCHMARKED AGAINST
Atlas -> Llama3
ATLAS was evaluated on 32-layer LLaMA3 with 10^225 configuration space.
Source finding

Source trail

GHSA-x2rj-828p-hx9m arXiv 2607.23478 arXiv arXiv arXiv GitHub

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…