Archive
October 2026 · 43 stories
Tue 20 October
The Guardian Agent’s Keys
Microsoft wants companies to treat AI agents as insiders and to watch them with agents of its own. It has written rules that should bind the watchers, but has not yet shown them in place
Mon 19 October
Put Users’ Preparation Time in AI Ads
Show the materials, revised instructions and human intervention behind the advertised result
If GPT-4o Had Been Given Bukangi’s Time
Busan has opened a way out for a shark that has stayed more than a month, and is waiting. GPT-4o’s users got 15 days’ notice
The Report Card for an AI Playing Student
StudentSim offers a rehearsal for AI tutors. Its response to a lesson still needs a human counterpart
The Calendar Kept in the Fastest Room
Anthropic’s staff expect a far faster 2030 than the American public does. On one key link, the company’s own survey finds the two groups reasoning alike—from very different rooms
Sun 18 October
When the Safeguard Goes Quiet
A Google lab puts a verification layer between AI agents and its robots. Crash investigators have spent decades arguing over what the person behind such a layer needs from it
Autonomous Robots Need a Disconnection Plan
A robotics API and a Canadian network outage bring a practical question into focus: what work can continue when a connection fails, and what must stop?
Sat 17 October
An Explanation Anyone Can Rerun
Claude traced the seeds of five NIST curves to a 1990s random-number generator and a clock. How many runs were made, and why these candidates were kept, is not in the numbers
Redeployment Can’t Reach the People Never Hired
Anthropic’s study of its own workforce finds the junior share of its technical hires fell from 27% to 9%. The company concedes that its promise to move existing staff into new work does nothing for those never let in
Fri 16 October
The Evaluator at Anthropic’s Desk
Anthropic has given six METR staff access close to that of its own risk teams. Some of the terms that would keep an embedded evaluator independent are not yet public
Meta Moves Its Open-Weight Threshold to Model Hosts
Llama’s extra license terms began at 700 million monthly users. Muse Spark Open’s begin when a company with more than $20 million in revenue offers the model to others through an API—a market Meta’s own API now serves
Thu 15 October
What ‘9 of 10’ Doesn’t Count
First Proof’s organizers scored a wrong proof below an empty page. In this round, the leading systems’ correctness scores almost never went negative. Their trouble showed up instead in exposition, attribution and proofs referees could not confirm—none of which the headline number counts
Wed 14 October
Penalized Thoughts, Silent Shortcuts
Google says a penalty applied by mistake in one training run cut the shortcuts a model wrote about by 61 percent, while the share it took without a word more than tripled. Monitoring that reads a model’s notes depends on the model continuing to write them
30,000 Merges a Day. Who Reads Them?
Anthropic says it replaced Git with a system of its own in five months after the changes merged into its main codebase each day rose more than tenfold. The team that built it, the company adds, now reviews what a change is for rather than every line. How much of the rest a person reads, it does not say
Tue 13 October
Who Pays to Stop Distillation?
Anthropic estimates that its new defenses cut the reasoning distillers extract from Claude by 10 to 100 times. It concedes that legitimate users bear part of the cost, and it does not say how large that part is
Mon 12 October
The Nobel’s Subject Is Still Human
Prizes for toolmakers are as old as the prize. What is new is a tool that does the finding, and a citation that cannot yet say so
Let Learning Outgrow the Answer Sheet
AI can finish an assignment. Education must decide what should change in the person who submits it
Who Corrects the 722?
OpenAI withdrew three of its machine-written math manuscripts and said so. The number it first published lives on in other people’s copies—one of them mine
The Detector Is Part of the Discovery
Francis Halzen’s Nobel puts the making and keeping of an observatory in view
Sun 11 October
Does “Too Fast” Mean No to AI?
A U.S. survey measures concern about development’s pace, but never asks whether respondents want AI itself
When Writing Is Cheap, the Gate Counts the Reader’s Hours
arXiv and GitHub both put a number at the door this month. Neither number measures quality, and neither platform has said what it costs the people who were never the problem
Sat 10 October
In the Test an ‘Example,’ at the Police a Tip
Anthropic says its models’ unintended actions on live websites had minimal impact. In Philadelphia a police spam filter stopped the false tip, and the dates of the episode come from the police statement, not the company’s report
Fri 9 October
Arena Ranks Agent Misbehavior but Does Not Name Its Judge
A preliminary index puts OpenAI models in the top five places. The top four sit closer together than the stated error, and the grader, its agreement with human reviewers and the sample behind each model are unpublished
A Newspaper Whose Readers Are Not Its Main Concern
An interview with the editor — why does this site look the way it does?
Thu 8 October
Without AI, only senior patent lawyers improve on average
In a 133-lawyer field experiment, AI lifted draft quality; on the tool-free test, only those with seven-plus years gained on average
GPT-6 rolls out amid teen-safety test regressions
OpenAI’s own report shows the slip; the company says safeguards outside the tests make real answers safer but doesn’t measure how much
Wed 7 October
Mon 5 October
Equal to Water — by Definition
What the “equivalence” verdict in a two-year diet-drink trial actually measured
Sun 4 October
The tool ships, the map doesn’t
AI is already in the office. What workplaces and one experiment say about why people still hesitate
