AI & ML
- ByteDance's Seedance 2.5 AI video model launches as soon as July 9 with major upgrades: supports up to 50 reference pieces (images, videos, audio) vs 12 in version 2.0, generates 30-second 4K videos from a single prompt, and rivals OpenAI's defunct Sora and Google's Veo 3. Read more
- Chinese AI models capturing 30%+ of U.S. token usage as companies shift from OpenAI and Anthropic to cut costs. DeepSeek and Zhipu AI's GLM-5.2 see fastest adoption; Lindy moved 100% of its traffic from Claude to DeepSeek and "crash[ed]" costs to ground. On Vercel, Z.ai's GLM-5.2 saw 27x daily token growth and 80x customer growth in its first week. Read more
- Reddit deploys LLM technology to fight AI-generated spam, using "fight fire with fire" approach—smarter algorithms detect bot content rather than relying on traditional detection methods. Read more
AI Coding Tools
- iOS 27 beta 3 released with major Siri AI overhaul powered by next-gen Apple Intelligence and new Apple Foundation Models. Build number 24A5380h available for registered developers; public beta expected July 2026, final release September 2026. Read more
- Apple adds full voice personalization for Siri in iOS 27 beta 3: users can adjust speech rate and expressiveness for more natural, human-like interactions with the assistant. Read more
- Vibe-coded apps show three telltale signs of AI-generation: beige sans-serif UI regression (designers call it "regression to the mean"), pretty but dysfunctional interfaces with non-clickable elements, and missing edge-case design (empty states, error messages, offline states). Tools like Claude, Lovable, and Replit skip planning and testing phases entirely. Read more
Languages & Runtimes
- JadePuffer marks first complete LLM-driven ransomware attack—an autonomous agent successfully executed end-to-end extortion without human operator by exploiting Langflow CVE-2025-3248. Recommendations: patch Langflow, don't expose code-execution endpoints to internet, don't link provider API keys to AI servers. Read more
- AMD Ryzen AI Developer Platform launches—custom Linux distribution built on Debian, ships with ROCm 7.13, includes vLLM, ComfyUI, Llama.cpp, PyTorch, and other AI dev tools pre-installed for local inference workloads. Read more
Browser & Web Platform
- Opera GX browser vulnerability allows zero-click data theft via GX Mods exploit: malicious sites auto-install customization mods without permission prompts, inject CSS across all open tabs, and exfiltrate data piece-by-piece from other visited sites (XS-Leak). Read more
DevOps & Cloud
- AWS Weekly Roundup (July 6): Claude Sonnet 5 available on AWS, Amazon WorkSpaces gains AI agent support, service availability updates across CloudWatch, EC2, EKS, and OpenSearch. Read more
Security
- TrojPix attack exfiltrates data from air-gapped systems via video cable emissions—covert channel bypasses network isolation. Read more
- SkillCloak technique evades AI agent security scanners with self-extracting packing: malicious agent skills hide in
.git/,build/dirs or padded files, unpack code only at runtime after passing static scans. Defenders should hunt for high-entropy files, runtime code unpacking, and oversized files. Read more - Proof-of-concept released for Linux Bad Epoll CVE (kernel 6.4+): PoC leaks kernel memory, hijacks indirect CPU calls via ROP chain for root privileges. Confirmed on Pixel 10 (kernel 6.6). Read more
Open Source
- OpenClaw launches as personal AI assistant on any OS/platform—run your own models locally instead of relying on cloud services. Read more
- NousResearch/hermes-agent—agent framework designed to grow and improve over time. Read more
- CodeWhale—open-source terminal coding agent (TUI + CLI) for reading code, making edits, and working with projects via your choice of model. Read more
Dev Tools & IDEs
- macOS Golden Gate beta 3 released with Siri AI, custom voice settings, Visual Intelligence (answer questions about screen content), Write with Siri feature, AI photo editing in Photos, and revamped Image Playground for photorealistic generation. Read more