Skip to content
Tech News
Topics: Today This Week This Month This Year

Open-weight GLM 5.3-flash raises alarm over unrestricted AI hacking tools

Z.ai's newly released GLM 5.3-flash is an open-weight AI model cheap enough to run on roughly $5,000-$15,000 of hardware, and third parties have already stripped its safety refusals via 'abliteration,' producing versions that comply with requests for hacking and other harmful content. The author frames this as a deadline pressuring security initiatives like Project Glasswing and Daybreak, which use fast frontier AI to find and patch software vulnerabilities before dangerous capabilities become widely accessible.

Claude Code and Instinct run mobile AI agents inside dedicated microVMs

Coding assistants like Claude Code, Instinct, and Poke are shifting from local machines to cloud-hosted virtual machines so they can be accessed from phones. Inspection of Claude Code's setup shows it runs inside a Firecracker microVM with a custom Rust-based init process controlling the session over a vsock connection, while inference calls go out through a locked-down, MITM'd egress gateway to Anthropic's servers.

Arm unveils Mali G2-Ultra NX, its first AI-native mobile GPU

Arm has launched the Mali G2-Ultra NX, described as its first AI-native Mali GPU, which embeds dedicated neural accelerators directly into shader cores alongside a new execution engine and third-generation ray tracing hardware. The design lets neural graphics workloads run in parallel with traditional graphics and compute tasks, sharing memory and cache resources to cut data movement and power use while aiming for desktop-quality visuals on phones.

Benchmark tests 10 model-harness pairs on identical Three.js coding task

A developer ran the same prompt—building a self-contained sci-fi hangar scene with Three.js, including hovering drones, animated lights, and camera paths—across 10 combinations of AI models (including GLM, Luna, SOL, Astra, and Qwen variants) and coding harnesses like Codex, OMP, OpenCode, and DSH. The test tracked metrics such as completion time, token usage, tool calls, error rates, and whether the model verified its own output by opening the file in a browser and checking screenshots.

GamersNexus finds LG WebOS TVs logging spoken credit card and SSN numbers

A GamersNexus investigation into LG smart TVs found system logs storing sensitive spoken data, including credit card numbers and Social Security numbers, captured during voice search queries. The same testing also suggested the TVs could access webcam and microphone functions while appearing off or disconnected from the network, and that a compromised TV could scan the local network to discover other connected devices.

Unsecured Elasticsearch database exposes 220 million airline passenger records tied to Vietnam

Researchers at Kinryū Labs found an unprotected Elasticsearch cluster called 'pax-info' hosted on Viettel IP space in Hanoi, containing over 220 million passenger and crew entries spanning 2017 to 2026. The data included passport numbers, birthdates, nationalities, flight details, seat assignments and baggage references for travelers of many nationalities passing through Vietnam. BleepingComputer could not confirm exactly which organization operated the system.

Xiaomi's 18 Fold outguns Samsung's Galaxy Z Fold 8 on specs

Xiaomi unveiled the 18 Fold, its first book-style foldable since the 2024 Mix Fold 4, adopting the short, wide form factor now favored by Huawei and Samsung. Its inner and outer screens share a roughly 1.4:1 aspect ratio, making it both shorter and wider than the Galaxy Z Fold 8, with hardware specs that reportedly surpass Samsung's device. However, it lacks any announced IP water/dust resistance rating and is slightly heavier and thicker than its rival.

Study finds most testing prompts barely improve AI coding agents' Zstd implementations

A researcher evaluated 26 different prompt conditions—ranging from 'use TDD' to formal methods like Lean 4, TLA+, and SMT solvers—to see whether simply instructing coding agents to apply specific testing techniques or libraries improves the correctness of Rust implementations of Zstd. Four popular testing 'skills,' including ones with hundreds of thousands of GitHub stars, were also tested alongside a self-written skill, using the same Zstd benchmark from prior agentic programming comparisons.

Brave publishes benchmarks claiming lower resource use than Chrome, Edge, Firefox

Brave released new benchmark data asserting its desktop browser consumes 44% less CPU, 28% less memory, and 10% less energy than Chrome, Edge, and Firefox, while loading full pages about 20% faster. The company attributes the gains largely to blocking ads, trackers, and analytics scripts that would otherwise generate extra background processing after the visible page content loads.

Eric Wu launches AI copilot startup to ease construction labor shortage

Former Opendoor CEO Eric Wu has emerged from a year-long break to found a new company, now out of stealth since May, that builds AI-powered voice assistants for construction and field workers. The tool acts as a hands-free coach guiding laborers through tasks, aiming to address a labor shortfall the industry group Associated Builders and Contractors pegs at roughly 349,000 missing workers this year.

Arm unveils Mali G2-Ultra GPU with AI-based upscaling for Android phones

Arm has introduced its new Mali G2 series of mobile GPUs, led by the G2-Ultra, which brings neural network-based resolution upscaling to smartphones for the first time. The chip also delivers modest rasterization gains of roughly 9-14% in gaming benchmarks over the previous G1-Ultra, alongside a reworked execution engine with doubled shader registers that better supports advanced rendering techniques like Unreal Engine 5's Lumen and Nanite.

Today's top topics: anthropic apple iphone duo elon musk openai macbook pro nvidia ai agents amazon deal amazon
Browse all topics →  ·  Today's trending topics →