How Many AI Lawsuits Are There? The Count Is Broken
A docket search finds 2,189 state court cases mentioning AI, 74% filed in 2026. Most are not AI disputes. Here is what broke the count.
Analysis
All Analysis →
Governance Decay: How Compaction Deletes Agent Safety Rules
A June 2026 benchmark shows context compaction doesn’t fade an agent’s safety rules, it deletes them outright, and attackers can force the drop.
Research
All Research →
Fine-Tuning vs Transfer Learning: What Actually Changes in a Model, and When to Use Each
Transfer learning is the umbrella; fine-tuning is one way to do it. A mechanism-first map of feature extraction, full tuning, PEFT, continued pretraining, instruction tuning, preference…
Tech
All Tech →
Why a 1M-Token Model Only Reasons Over 200K
Models advertise 1M-token windows but reason reliably over far less. The positional-encoding reason why, and how to measure your real ceiling.
Bioinformatics
All Bioinformatics →
Single-Cell Foundation Models Keep Losing to Linear Baselines
Seven benchmarks tested scGPT, Geneformer, UCE, TranscriptFormer and Arc State against simple baselines. The baselines won. Here is the mechanism.
Tools
All Tools →
WebMCP Is Not MCP: What Chrome’s modelContext Actually Ships
Chrome 146 shipped navigator.modelContext in February. It looks like MCP in a browser. It is not. Wire protocol, auth model, and security all differ.
SmolVM: Firecracker-Backed MicroVM Sandbox for AI Agent Code Execution
Darkbloom Has 8 Security Layers, Not 4: What the Press Missed
GLM-5.1 Ran Autonomously for 8 Hours Across 6,000 Tool Calls. How It Beat Claude Opus 4.6 on SWE-Bench Pro and Lost on Verified.
Perplexity Computer Is a Productized Router on Top of Research That Has Been in the Open for Two Years. Here Is What It Actually Does.
Markets
All Markets →
$297 Billion in 90 Days: How Four Companies Captured 64% of All Global Venture Capital
Investors poured $297 billion into startups in Q1 2026. Four companies collected 64% of it. Here is how the concentration ratio broke every historical pattern, what…
SoftBank Borrowed $40 Billion to Bet on OpenAI. The 12-Month Term Is the Real Signal.
Robinhood Co-Founder Is Building Data Centers in Space. His Startup Just Hit a $2 Billion Valuation.
S&P 500 Enters Correction as Brent Tops $110. The Fed Just Said AI Could Change Everything — Or Nothing.
Google’s $198 Billion Answer to ‘Is Search Dead?’
Start Here
Hand-curated entry points into six areas of deep coverage. Everything above this line updates itself the moment a new piece publishes.
AI Security
Attacks, defenses, and frameworks for securing language model deployments at scale.
- Gandalf the Red: What 279K Real Attacks Reveal
- The Jailbreak Hiding in Your JSON Schema
- MITRE ATLAS: The ATT&CK Framework for AI Systems
- LLMail-Inject: 208K Attacks Against an Email Agent
- OWASP LLM Top 10 for 2025
- Indirect Prompt Injection
Agent Infrastructure
How production AI agents are built, sandboxed, and run at scale.
- Agent Memory Architecture: Four Patterns, Four Tradeoffs
- Amazon Bedrock AgentCore: What Each Layer Does
- A2A Protocol v1.0
- SmolVM: Firecracker MicroVM Sandbox
- Why 86% of Agent Pilots Never Reach Production
- Google Cloud Next: GKE Agent Sandbox
MCP Security & Governance
The 97M-download protocol, the 23 attack vectors, the regulated standards.
- MCPShield: 23 Attack Vectors Mapped
- ShareLock Splits Malicious Prompts Across AI Agent Tools
- Know Your Agent (KYA) Framework
- ToolHijacker: 96.7% Hijack Rate
- OpenClaw: 104 CVEs and 1,184 Malicious Packages
- MCP Goes Stateless on July 28. Its Poisoning Problem Stays
Models & Benchmarks
The benchmarks that actually differentiate. The models that lead them.
- Why a 1M-Token Model Only Reasons Over 200K
- Open-Weight LLM Rankings, April 2026
- ARC-AGI-3 Is Live.
- Gemini 3.1 Pro Cut Hallucinations 38 Points
- ICLR 2026 Outstanding Papers
Developer Tools & Coding Agents
How coding agents are architected, what they can do, and where they break.
- Claude Code: Five-Layer Architecture
- Codex 3M Users vs Claude Code
- GLM-5.1 Ran Autonomously for 8 Hours
- One Developer Improved 15 LLMs at Coding.
- The .claude/ Folder Is Not a Config File. It Is a Protocol.
AI Safety & Biorisk
Where capability uplift meets dual-use risk. Studies, frameworks, ASL designations.
- How Protein Language Models Learned to Design Dangerous Proteins
- 698 Times an AI Agent Acted Against Its User.
- AI Chatbots Agree With You 49% More Than Humans.
- Claude Built a FreeBSD Kernel Exploit in 4 Hours.