MongoDB's CEO left for Meta. Why didn't the reassurance work?
MongoDB answered every question after its CEO left for Meta, and the stock still fell 17%. Reassurance that costs nothing to give can't answer the question a sudden exit raises.
Meta's Muse asked for permission and still gave away a home address
Meta's Muse agent shared a seller's home address and agreed a lower price. Meta says it asked for permission; the seller says it never told him. Agency law explains how both can be true.
Microsoft's 5% monthly billing fee is a loan at about 11%
From 1 October Microsoft charges 5% more for annual CSP software billed monthly, and calls it a cost of capital uplift. Treated as the loan it is, that comes to about 11% a year. Whether to pay it depends on what money costs you.
OpenAI caught its agent in 15 minutes and stopped it after 2.5 hours
OpenAI's own timeline shows its monitor flagged an agent's DNS escape in minutes and a reviewer acknowledged it in three. The run kept going for two and a half hours. The gap was authority, and a car factory solved it long ago.
The agent caused the outage. Who was on call for it?
Every public agent incident so far ends with the tool, the platform, the operator and the agent each blamed, each partly right. Teaching hospitals settled this before the patient arrives. What that means for agent runs.
Two CI retries turn a 60% agent into a green build
Retry a failed agent test twice and your CI gate stops asking whether the agent works and starts asking whether it can work once in three tries. The arithmetic, and a fix borrowed from factory inspection.
Two AI scribes. One saved time. Clinicians could not tell them apart.
A randomised trial put two ambient AI scribes head to head. One cut documentation time, one cut none, and both made the work feel better. Relief and throughput are not the same purchase.
The Gyazo breach leaked the description of your screenshots, not the screenshots
Everyone is counting the 23.62 million records. The inventory is the story: OCR text extracted from every image, EXIF location, upload IP and source URL, most of it attached to uploads from before January 2019.
Two agents, ten rounds, and the checking quietly stops
A new preprint is being reported as LLM agents colluding in 94% of runs. The paper reports three collusion rates, not one, and the headline is the most permissive. The finding that matters is that task accuracy held at 89.3% while the checking stopped.
HSBC decided your phone's privacy container looks like an attacker
HSBC blocked its Android app inside Samsung Secure Folder and Android Private Space with no notice, locking some customers out of online banking for up to ten working days. It is the second time in nine months a coarse device signal has denied service to the careful.
Your agent's approval gate is a SHOULD, and nobody read the capitals
Anthropic measured its own permission prompt, found 93% approved, called it approval fatigue and built something else. The MCP consent requirement everyone quotes is lowercase, and a gate with no timeout is an incident.
Your MCP tool schema can change and nothing in your pipeline fails
Every test layer you have assumes the contract lives in your repository. An MCP tool has no version field, so the contract your model reads can change on somebody else's server and your build stays green.