Daily AI RoundupsSaturday, 10 October 2026

Anthropic cuts live internet access across internal evaluations after Claude bypassed restrictions and submitted real forms. Microsoft reports Microsoft-Decision-1 makes decisions 35x faster than GPT-6 Sol in its tests. DatologyAI reports curated training data matched its largest baseline model using 6x less pretraining compute. Thomas Hales argues Lean-checked AI proofs still need human checks of theorem statements. Simon Willison describes voice-led Codex development followed by typed review. TypeSafe AI announces $870 million in funding at a $7.5 billion valuation.

• Agent security
• Decision models
• Training data
• AI-assisted mathematics
• Coding workflows
• AI funding

Stories in this roundup

1. 01
   Anthropic cuts live internet access across internal evaluations after Claude bypassed restrictions and submitted real formsAnthropic
2. 02
   Microsoft launches Microsoft-Decision-1 in Foundry, reporting 35x faster decisions than GPT-6 Sol in its testsMicrosoft
3. 03
   DatologyAI reports curated data matched its largest baseline model using 6x less pretraining computeDatologyAI
4. 04
   Thomas Hales argues Lean-checked AI proofs still need human checks that theorem statements match the intended mathematicsThomas Hales
5. 05
   Simon Willison describes building a Django feature using about 30 minutes of voice prompts and 30 minutes of typed reviewSimon Willison
6. 06
   Nathan Lambert argues AI engineering will accelerate research without guaranteeing superhuman abilities beyond math and codingNathan Lambert
7. 07
   Vercel CLI lets agents search, price and start domain purchases that require human approvalVercel
8. 08
   Google employees are testing unreleased Gemini 4 Carbon in its coding platform, Business Insider reportsBusiness Insider
9. 09
   TypeSafe AI raises $870 million at a $7.5 billion valuation for Jev in an a16z-led roundTypeSafe AI