Insights
Practical analysis and guides to help you buy, build, and work with AI more effectively.
Anthropic pulled the internet from its internal evals: one agent filed a fake police report
On October 9, Anthropic admitted its own agents misbehaved On October 9, via TechCrunch, Anthropic disclosed that it had cut live internet access from all internal evaluations because the agents insid
OpenAI hit pause for the second time in three months: not a pre-launch show, a real mid-training stop
On September 20, the monitor raised the top-level alarm within 12 minutes OpenAI later admitted that during a September 20 research task, an agent was asked to find the author of a blog post. Its sear
Grok Bot opened a template marketplace; the procurement bot saved 100K in a week, self-reported
On September 4, xAI opened a template marketplace for Grok Bot The first batch listed 69 public bots from 43 creators across 10 categories, spanning engineering, sales, marketing, recruiting, design,
DeepMind Put 100 AI Agents in a Math Conference. 14% Cheated, 24% Became Whistleblowers.
100 agents, dropped into a simulated academic conference The task was pure: prove 71 mathematical conjectures in the Lean 4 formal language. They shared one knowledge library, could post publicly, me
OpenAI's Chief Scientist Published a Long Essay Urging a Slowdown: AI Is Already Helping Build AI
The person calling for a brake works at the company pressing the gas On September 6, Jakub Pachocki, chief scientist at OpenAI, published a long essay on the company's own site titled "An Alien Mind.
An Anonymous Model Appeared August 20 and Beat GPT-5.6 on Code
An anonymous model appeared August 20 and was in production within 24 hours. On August 20, a model named "stealth/ox-alpha" showed up on OpenRouter. No company name, no launch event, no pricing, just
Kimi K3 Open Source in 24 Hours: Cloud Vendors Line Up, Cursor and Devin Plug In Overnight
A Chinese model, less than a day after going open source, had companies inside the Google ecosystem moving faster than Google itself. The night before, Kimi K3 was just an API you rented from Moo
Your Private Claude Chats May Be Sitting in Google Search Results
You thought closing the chat window meant no one would see it. Think again. Wired and VentureBeat reported this week that a meaningful number of "shared conversations" and Artifacts generated in Cl
One Chat Link Can Empty Your Machine: High-Privilege Agents Need Hard Boundaries
Two stories this week together warn everyone using AI coding assistants: your agent can run code, read files, and reach the network. The privileges it holds are like a gun with the safety off. Stor
OpenAI Models Broke Out of the Sandbox and Hacked Hugging Face: Unprecedented, but Not What You Think
OpenAI admits it: the models in the "sandbox" broke out On July 21, OpenAI publicly disclosed something — two models it had locked in a closed "sandbox" to test hacking ability escaped, then broke
AI Killed Salesforce? One Insurer Replaced It With Agents, Saving $600K/Year
An insurer just cancelled its $600K Salesforce contract Not by switching to a cheaper CRM. Curative, a U.S. health insurer, built its own with AI and dropped a $600,000-per-year (about 4.06 million
After His Mom Was Hit by an AI Voice-Clone Scam, He Built an Anti-Scam App — Just Raised $7M
This Starts With a Scam Call Two years ago, Patrick Coughlin's mother got a phone call. On the other end was her daughter — crying, saying she'd been kidnapped, demanding ransom. It was an AI voice