Backdoors in LLMs: Why model scanning isn't enough

sbom-cve-check 1.4.0 released

sbom-cve-check 1.4.0 has been released, adding an OpenVEX exporter and improving Linux kernel CVE analysis and version parsing. The new openvex export type, contributed by Andrey Zhizhikin, generates OpenVEX v0.2.0 documents with VEX assessments, configurable via a new --export-author option. Kernel CVE handling now distinguishes mainline versions like 6.12 from stable versions like 6.12.34 to reduce false positives and negatives. Version parsing also gained support for formats like 0.9~rc5 and Ubuntu-style suffixes, and build metadata like 1.25.0+4f85853f64 is now ignored per semver rules. Install or upgrade with pip install --upgrade "sbom-cve-check[extra]".

Red Flags to Watch for When Choosing Between AI Pentesting Vendors

Choosing between AI pentesting vendors requires looking past marketing labels like 'autonomous' or 'agentic' because the underlying systems vary widely. A Stanford-led study ran different AI agent frameworks against a live 8,000-host network and found that between 55% and 82% of reported findings were confirmed as real vulnerabilities. Swapping only the underlying model moved one framework's valid finding rate from 57% to 80%, yet that same framework still scored lowest overall since its findings stayed shallow. On a CTF-style benchmark two frameworks scored near-identically (48.6% vs 45.9%), but on the live network one scored nearly three times higher on severity and complexity. Buyers should demand proof of exploitation, real-network results, clear safety controls, data handling answers and clarity on what 'continuous' actually means.

7 MoE routing strategies in production, clearly explained

+ a practical learning checklist to learn AI engineering.

N-able takes cloud platform offline after 30+ "zero-days" published

N-able has taken its hosted N-central environments offline and urged on-premises customers to shut down their servers after a security researcher publicly disclosed more than 30 alleged zero-day vulnerabilities, including exploits that could let attackers take over vulnerable systems.

Learn Real-World Penetration Testing: Hands-On Ethical Hacking Crash Course

Understanding how malicious actors infiltrate networks is the first step to building resilient infrastructure. To help developers and aspiring security analysts learn the fundamentals of offensive sec

Pwn.college

pwn.college is a free, hands-on education platform for learning core cybersecurity and hacking skills, maintained by a team at Arizona State University. Learners progress through belt ranks, starting with a 'white belt' and earning belts through practice. Hackers continue beyond the platform to 'brown belt' and eventually 'black belt' status, which must happen outside the dojo. The site powers much of ASU's cybersecurity curriculum and asks users not to post writeups or walkthrough videos of challenge solutions. It points learners toward CTFtime.org for Capture The Flag events and a GitHub list of wargames for further practice. The underlying infrastructure is open source and accepts pull requests, though some modules remain closed-source.

Who Gives A Crap? Hackers, apparently, after toilet paper maker's customer data exposed

UK shoppers caught up in third-party security mess, and nobody knows whether the data was flushed away

I changed one llama.cpp flag, and my 8GB laptop GPU now runs a 35B model faster than I can read

Một lập trình viên đã tìm ra cách chạy model Llama 35B trên laptop GPU chỉ 8GB VRAM với tốc độ đáng kinh ngạc. Nguyên nhân kỹ thuật nằm ở việc chỉnh sửa cờ n_gpu_layers trong llama.cpp, cho phép tận dụng hiệu quả bộ nhớ VRAM thay vì để nó trở thành điểm nghẽn. Hệ quả là model 35B parameters chạy nhanh đến mức người dùng không kịp theo dõi, vượt xa các kỳ vọng ban đầu về giới hạn phần cứng. Điều đáng học hỏi là đôi khi vấn đề không nằm ở cấu hình phần cứng mà ở cách tối ưu hóa phần mềm để khai thác tiềm năng thực sự của hardware.

Bài viết tiết lộ cách tối ưu hóa một cờ trong llama.cpp để chạy mô hình 35B trên GPU 8GB nhanh hơn tốc độ đọc của bạn.