AI digest: Big moves and power plays
Anthropic debuts security-focused Mythos model, Google ships offline dictation, and the usual chip wars heat up.
The week’s biggest stories show AI companies making serious enterprise plays whilst the hardware wars intensify.
Anthropic launches Mythos for cybersecurity
Anthropic debuted Mythos, a powerful new AI model specifically designed for defensive cybersecurity work. Only a handful of high-profile companies get access to the preview. This feels like Anthropic positioning itself as the “responsible AI for serious enterprise” player, especially after their recent stance against military applications.
Google goes offline with AI dictation
Google quietly released an offline-first dictation app using Gemma models to compete with apps like Wispr Flow. Running AI models entirely on-device for dictation is genuinely useful, though it’s odd they launched so quietly. Privacy-conscious users will appreciate not sending every word to the cloud.
Anthropic signs massive compute deal
Anthropic inked a multi-gigawatt TPU deal with Google and Broadcom coming online in 2027, as their revenue run rate hits $30 billion. The scale is staggering and shows how compute requirements are exploding faster than anyone predicted. Google’s clearly betting big on Anthropic challenging OpenAI.
Tiny Arcee AI punches above its weight
26-person startup Arcee built a high-performing open source LLM that’s gaining traction with OpenClaw users. It’s refreshing to see small teams compete with the giants through clever engineering rather than just burning cash. Open source might be the only way to prevent complete Big Tech dominance.
Intel joins Musk’s Terafab project
Intel signed onto Elon Musk’s Terafab semiconductor project alongside SpaceX and Tesla for a new Texas fab. The scope of Intel’s contribution remains unclear, but anything that reduces chip dependence on TSMC seems strategically smart.