AI infrastructure and macOS productivity tools lead today's launches.
Anthropic drops a bombshell accusing Chinese AI labs of industrial-scale model theft, IBM takes a massive hit from Claude Code's COBOL capabilities, and OpenAI abandons its own benchmark.
Anthropic reveals 24,000+ fake accounts and 16M+ exchanges used to systematically extract Claude's capabilities. Three Chinese labs named — DeepSeek, Moonshot AI, and MiniMax.
Evidence shows coordinated model distillation at industrial scale. The report drops as Congress debates AI chip export policy — timing that hasn't gone unnoticed.
Quick hits — local models, benchmarks, and hardware.
ByteDance challenges image gen leaders, and local music generation takes a big step forward.
A security wake-up call for self-hosted *arr stacks, plus Mac utilities and self-hosted tools.
Community discovers that Huntarr, a popular *arr stack automation tool, was broadcasting Sonarr/Radarr API keys and passwords in plaintext to anyone on the network.
The tool was entirely "vibe-coded" — AI-generated without security review. GitHub repo since taken down. A cautionary tale for self-hosted AI-generated software.