0. AI is forcing us to pay off 40 years of technical debt (spectrum.ieee.org)
1304 points · 749 comments · by pseudolus
AI-assisted research exposed the PHP2Shell vulnerability in legacy PHP code before exploitation, highlighting cybersecurity’s growing technical debt and the need for better-funded defenders, automated formal verification, and prompt software updates. [src]
1. Residential proxies detected on White House network (layer3intel.com)
1268 points · 436 comments · by Rasbora
Layer3 Intel observed 236 residential-proxy exits across six White House network addresses over 26 days, suggesting a potentially compromised device, though the data identifies network activity—not the device, user, or cause—and researchers offered raw logs to relevant agencies. [src]
2. Ask HN: What's your current AI agentic setup?
564 points · 695 comments · by eggbrain
An Ask HN post seeks current information on users’ AI subscriptions, models, coding harnesses, orchestration tools, broader agent platforms, and the reasons behind their workflows and tool choices. [src]
3. US voters roundly reject data centers in ballot measures (semafor.com)
350 points · 756 comments · by justonepost2
US voters roundly reject data centers in ballot measures: Title: US voters roundly reject data centers in ballot measures
URL Source: https://www [src]
4. North Korean fake job interview led to $6.3B Binance heist (theregister.com)
657 points · 385 comments · by smalltech
North Korean hackers used a fake job interview and malware-laced coding assignment to compromise Binance engineers’ devices, exploit internal tools and credentials, and execute the exchange’s record $6.3 billion cryptocurrency theft on October 29. [src]
5. Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
734 points · 201 comments · by jameslk
Google’s Gemini personal accounts may use saved chats to improve AI models and have some reviewed by humans—even for subscribers—unless users disable Keep Activity or use temporary chats, while reviewed conversations may be retained for up to three years. [src]
6. Nvidia RTX Spark encrypted firmware slows inference speed to 35% on Linux (mindstudio.ai)
617 points · 291 comments · by novaRom
Linux runs on Nvidia RTX Spark laptops, but encrypted Windows-only power-management firmware locks the GPU to about 780 MHz when plugged in, reducing inference from 65.4 to 13.9 tokens per second; WSL maintains near-Windows performance. [src]
7. A thousand fewer paths to root (openai.com)
488 points · 365 comments · by dangelosaurus
OpenAI’s automated Linux kernel security pipeline identified 2,308 candidate flaws, reproduced 1,724, developed 948 local privilege-escalation exploits, and generated patches for maintainers, demonstrating both the scaling potential and validation challenges of AI-driven vulnerability research. [src]
8. Kimi K3.1 Pro model weights are out (huggingface.co)
640 points · 191 comments · by abrarthought
Moonshot AI has released the open-weight Kimi K3.1 Pro, a 2.8-trillion-parameter multimodal Mixture-of-Experts model with a 1-million-token context window, API access, and support for vLLM, SGLang, and TokenSpeed. [src]
9. AMD expects to outsell Intel in Q4 for the first time in history (tomshardware.com)
593 points · 224 comments · by rbanffy
AMD forecasts record Q4 revenue of about $17 billion, potentially surpassing Intel for the first time, as booming demand for Epyc processors and Instinct AI accelerators offsets declines in its client and gaming businesses. [src]
Brought to you by ALCAZAR. Protect what matters.