The Cheapest Inference Is a Classifier, Not a Chat Model, in a Bit
Microsoft-Decision-1 returns a label instead of a paragraph and costs $0.042 per million input tokens. A two minute read on decision models.
One Prompt Owned Every Agent in the AWS Region, in a Bit
One prompt to a single AWS Bedrock AgentCore agent returned the credentials for every agent in the account and region. A two minute read.
Confidential Filings Are Becoming the Standard Path for AI Power
Meta's 4,500 megawatt data center power demand was confirmed in testimony, and ten new gas units now hang on a largely confidential record.
GlobalFoundries Is Building the AI Chip Bottleneck a US Home, in a Bite
GlobalFoundries will make silicon interposers for TSMC CoWoS packaging in New York, the first US source for an AI hardware bottleneck.
Your vLLM Cache Server Answers an Unauthenticated Root Shell, in a Bite
LMCache CVE-2026-105192 is an unauthenticated root shell in the cache layer that speeds up vLLM, and no fixed version exists.
Windows Now Contains Agents at the Operating System, in a Bite
Microsoft Execution Containers are generally available, putting agent containment inside Windows 11.
750 Megawatts of AI Capacity Is Already Built and Idle, in a Bite
The National Compute Grid says about 750 megawatts of AI capacity is already built and mostly idle.
The Token Bill Became a Management Line, in a Bite
Microsoft and Meta are capping internal Claude use as AI token costs outrun projections. Here is the full analysis.
One Agent Handed Another Agent a Hostile Instruction, in a Bite
Five organizations disclosed the same MCP prompt injection flaw in five months. The fix is a redirect policy and an allow-list, not a model change.
Firmus Delivered 42 Megawatts of a 1.6 Gigawatt Plan, in a Bite
Project Southgate promised 1.6 gigawatts and delivered 42 megawatts. The Firmus and CDC partnership behind the Australian buildout has ended.