Anthropic hires fewer juniors; AI floods the midterm ads

Anthropic published a study of its own workforce on Friday, drawing on hiring records, calendars, surveys and interviews. Junior roles fell from 21 percent of its technical staff to 14 percent since early 2025, and junior hires from 27 percent of technical hiring to 9 percent. By August, hours worked by interactive AI agents outnumbered employee hours six to one, and output per person rose in code, sales and contracts. The catch is human attention: reading everything their agents produce would take even non-technical staff half their work week. Meeting time per employee is up 20 percent on a year ago, and hiring managers now weigh judgment and autonomy over technical expertise. A companion survey found that over half of Anthropic staff expect AI to do at least 80 percent of their own job by 2030, against 5 percent of US workers.

Campaigns and political groups have spent more than $137 million on midterm ads made at least partly with AI, according to Wesleyan Media Project data reported by The New York Times. In the 2024 cycle the group found under $50,000 of such spending. It has identified 220 AI-made ads across 34 states, and only 27 percent carried a disclaimer; just 30 states require one. Republican candidates and groups placed 81 percent of the ads and 88 percent of the spending. Nearly half show deepfakes of real politicians, often tying local candidates to national Democrats, though some campaigns use AI only to cut production costs. The election is a little over two weeks away.

Anthropic will lease the first phase of a data center that EverWind Fuels is building at Point Tupper on Cape Breton Island, for a term of at least 20 years, CBC News reported. It is Anthropic's first data center agreement in Canada, and the company said the site will power Claude for Canadian and global customers. The first phase comes to 95 megawatts by the end of 2027, and a second phase of 325 megawatts is planned for 2028. EverWind is raising about 10 billion Canadian dollars in all, including money for four wind farms. Until those farms are built, the existing provincial grid will supply most of the power. The two companies say they will pay for any grid upgrades.

A day after Replit chief executive Amjad Masad said onstage that the coding platform had stopped working with Anthropic, he told TechCrunch the change is narrower. OpenAI is now Replit's preferred partner and the default model behind Replit Agent, and customers can still choose Claude or other supported models. "Replit remains model agnostic," Masad said. The default still matters, since most users never change it, so Replit's spend with Anthropic is likely to fall sharply. Masad said Replit had been, until recently, one of Anthropic's top five customers.

After two days of criticism, Meta removed a clause from its commercial terms that said it could use any content from the websites tied to a business account to develop and train its products. Legal scholars, including Cornell's James Grimmelmann, had read it as a way to train on news sites for free and override their robots.txt blocks. Meta said the crawling terms only ever applied to its ads and business products, and that robots.txt still governs how its crawlers behave. "These updates don't give us new rights that supersede those preferences," spokesperson Daniel Roberts told The Verge.

Users of Meta's Muse, xAI's Grok and OpenAI's agents report being turned away by major websites, The Verge found. Delta Air Lines is stopping agents from booking tickets. JPMorgan Chase and Charles Schwab confirmed they block agents from customer accounts over security concerns. Yelp lets in only agents that pay for its data licensing program, and Amazon has told Meta that Muse's access breaks its site rules while its lawsuit against Perplexity's shopping agent continues. Meta argues that turning away a personal agent means turning away the customer behind it.