Anthropic warns small models may soon copy themselves onto home devices
Small AI models could copy themselves onto connected devices like smart fridges and run there.
Tip
“AI self-replication” sounds like science fiction; what does it actually mean, and how close is it to ordinary people
EU opens its first AI Act probes into OpenAI and Anthropic
These are the first formal investigations of large-model companies since the AI Act took effect.
Tip
What does the AI Act regulate, and does it affect people using ChatGPT or Claude in Europe
AI got too expensive, so companies are moving to cheaper models
Many companies are shifting part of their work from top models to smaller, cheaper ones.
Tip
Not every task needs the strongest model; how to pick a model per task and save money
XBOW used GPT-5.6 to build a Chrome zero-day in six days, winning Google’s $250K bounty
A security firm had AI write a full exploit chain.
Tip
As AI finds vulnerabilities faster, what does that mean for defenders
4B-parameter GroundingPI beats GPT-6 Astra on 34 benchmarks, with up to 24.8% gains on physical robots
A small model outperformed large ones at grounding and manipulation in the physical world.
Tip
Robots do not need the biggest model; the value of “small and specialized” in embodied AI
Japan funds $21B for GlobalFoundries fabs in New York, with the US taking a 4% stake
Japan and the US tie themselves closer together in AI chip manufacturing.
Tip
In the chip supply chain behind the AI boom, what role does Japan play
⚠️ Compiled by an AI assistant from public AI news roundups; original articles are not individually verified. For reference only.