← Back to archive
Cursor just bought Graphite—here's why it matters

Image via One Useful Thing

Monday, December 22, 2025

Cursor just bought Graphite—here's why it matters

Cursor just scooped up Graphite to merge code generation with AI-powered review (smart consolidation), while Anthropic dropped Bloom, an open source tool for scaled behavioral evaluations that could reshape how we stress-test these models. Meanwhile, SoftBank is in full sprint mode to hit its $22.5 billion OpenAI commitment by year-end through some creative asset shuffling (yikes), and researchers are discovering that AI agents are actually outperforming cybersecurity professionals in real penetration testing scenarios (wild). If AI agents are already beating human hackers, what's your move?

Top Stories

1
Cursor Acquires Graphite

TechCrunch

Cursor acquires Graphite to merge AI code generation with specialized code review tools, streamlining the entire developer workflow and addressing the quality issues inherent in AI-generated code.

ai-codingcursorgraphitem&a
2
Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing

arXiv

A new arXiv study compares AI agents directly to human cybersecurity experts in penetration testing, revealing how close autonomous systems are to matching professional-level security testing capabilities.

ai-agentscybersecuritypenetration-testingautonomous-systems
3
Introducing Bloom: An Open Source Tool For Automated Behavioral Evaluations

Anthropic

Anthropic released Bloom, an open-source tool that automatically generates behavioral evaluations for AI models to accelerate alignment research and address evaluation obsolescence. This addresses a critical bottleneck in AI safety by enabling scalable, configurable, and reproducible evaluation of frontier models in days rather than traditional timeframes.

anthropicalignmentopen-sourcesafety
4
Experiment Diary

Google Docs

Researchers document a series of RL training experiments on an LLM with progressively refined reward functions and constraints, showing an iterative approach to optimizing model behavior through reinforcement learning.

llmreinforcement-learningmodel-trainingreward-functions
5
SoftBank Races to Fulfill $22.5 Billion Funding Commitment to OpenAI by Year-End

Yahoo Finance

SoftBank is racing to close a $22.5 billion OpenAI funding commitment through asset sales and strategic liquidity moves, betting heavily on AI as OpenAI's valuation tripled since the initial investment agreement.

openaisoftbankfundingventure-capital

Keep Reading

Industry Voices

Enjoyed this issue?

Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.