AI in Technology

Anthropic’s Claude now supports system prompts to improve AI behavior control, marking a significant update in AI alignment and safety.

Analysis of how leading tech companies adapt to platform shifts in AI, highlighting lessons from history and current industry moves.

New research indicates AI systems now access vastly larger working memory capacities than humans, raising questions about AI capabilities and limits.

Anthropic has made auto mode the default for Claude Code sessions on select plans, enabling autonomous coding with safety classifiers—impacting AI safety and productivity.

A developer reports using AI-assisted auto-research with Codex to optimize kernel code, resulting in a 232-fold speed increase. Details are emerging.

Google introduces homomorphic encryption to make private AI more practical, enabling secure data processing without exposing sensitive information.

Analysis of recent user feedback suggests Opus 5’s usability issues are making it harder for developers, with ongoing investigations into the causes.

DeepSeek has released a developer preview of its new Harness platform, enabling early access for AI developers to test and integrate.

Google has launched Gemini 3.7 Flash, a new AI model update aimed at improving performance and efficiency, with details on capabilities and future plans still emerging.

Hugging Face’s large-scale reproduction challenge tested over 2,200 ICML 2026 papers, confirming thousands of claims but also revealing many contested results.