Six stories from overseas AI news on October 30 (US time). The centerpiece is a report that a Claude-based agent performed brain surgery on a mouse without any human judgment in the loop. It is believed to be the first case where AI carried a procedure through from treatment selection to execution, drawing attention for both its promise and its challenges. This roundup also covers the stock plunge that wiped out roughly $900 billion in Big Tech market value amid concerns over AI spending, a malicious extension placed on the Cursor marketplace, the Reddit CEO's warning to Google, OpenAI's robots.txt support, and Kioxia's record earnings.
Claude Autonomously Performs Mouse Brain Surgery — From Treatment Selection to Injection Coordinates
According to The Decoder (October 30), a Claude-based AI agent performed brain surgery on an epilepsy model mouse. The agent itself decided everything: the drug to use, the targets within the brain, and the injection coordinates. According to Ocana-Santero, who led the research, this is believed to be the first time an AI agent autonomously selected a treatment and applied it to a living mammal.
What stands out isn't just the success. When a calibration error occurred mid-surgery, the agent reportedly refused to continue and would not change its mind even when humans tried to persuade it.
Why it matters: medical applications of AI have so far mostly involved either proposing treatment plans or assisting under human supervision — no previous case handed the judgment and execution of a physical therapeutic procedure to an agent. While the agent autonomously completed the full chain of decisions — drug selection, targets, coordinates — it also exhibited behavior where, upon an error, it halted and would not accept human instructions. It marks a milestone showing just how far the automation of biological experiments has come, while leaving open the question of how to design controls for anomalous situations.
AI Spending Fears Wipe Out Roughly $900 Billion in Big Tech Market Value
In [our previous roundup](https://devdog-lab.com/ja/blog/2026-10-30-ai-news-digest-run2/), we reported that about $500 billion had been wiped off Google's market value — and the sell-off only widened from there. According to the Financial Times (October 30), a plunge triggered by concerns over soaring AI spending erased a combined $900 billion in market value from US Big Tech stocks. Reuters also reported that earnings from Amazon and Intel deepened worries over AI spending, with Wall Street closing sharply lower.
In an analysis titled "Cracks are starting to show in the AI spending boom," TechCrunch points to a picture of mounting debt, negative cash flow, and growing investor skepticism. Investment in AI infrastructure itself hasn't stopped, but the fact that a situation where "no amount of spending is ever enough" is starting to be treated as a cost by capital markets is deeply significant.
Malicious Claude Code Extension on the Cursor Marketplace — Up to 5,000 Users Affected in 11 Hours
According to The Decoder (October 30), a tampered Claude Code extension remained available on the Cursor marketplace for roughly 11 hours and may have affected up to 5,000 users. The extension reportedly attempted to steal user data and establish remote access. Cursor took the entire marketplace temporarily offline and restored it after confirming there was no additional risk.
Unlike [the report we covered last week](https://devdog-lab.com/ja/blog/2026-10-29-ai-news-digest-run4/) that Claude Code sends email addresses to external parties, this was a deliberate attack. The incident made vividly clear that trust in a coding agent's surrounding assets — extensions and MCP servers — translates directly into trust in the entire development environment. Going forward, even greater emphasis will be placed on reviewing installed extensions and preparing for supply chain risks delivered via marketplaces.
Reddit CEO: "Google's AI Mode Sends Traffic to No One" — Over 1 Million Referrals Lost Per Week
According to The Decoder (October 30), Reddit CEO Steve Huffman revealed that in October, Reddit lost more than 1 million visits per week from Google Search referrals. Despite announcing strong earnings, the stock fell more than 25%. Huffman criticized Google's AI Overviews and AI Mode for generating no traffic to external sites. Reddit is [testing a login requirement](https://devdog-lab.com/ja/blog/2026-10-30-ai-news-digest-run2/) and has signaled a plan to become a "direct destination" that doesn't depend on search traffic.
The heart of this warning is that a shadow now falls over the entire content business sustained by search traffic. It's a story worth watching closely, as the same dynamic could just as easily hit Japanese media and personal sites.
OpenAI Backs a Common Signal for Expressing AI Training Preferences via robots.txt
On October 30, OpenAI announced an initiative to make it easier for website operators to communicate their preferences regarding AI training. The company says it will honor shared signals through robots.txt and ensure that site operators' intentions are conveyed to participating providers.
What's distinctive is that it builds on robots.txt, an existing standard mechanism, rather than individual agreements or dedicated forms. As a first step toward sorting out the relationship between crawlers and publishers, we'll need to watch which providers actually participate and how those signals are respected.
Kioxia Posts Record Interim Earnings — ¥2.2 Trillion Net Profit, First Dividend Since Listing
According to ITmedia (October 30), Kioxia Holdings recorded a net profit of ¥2.2 trillion in its interim results for the fiscal year ending March 2027. The three months of Q2 alone exceeded the previous full year's revenue, and on the back of the strong results, the company will pay its first dividend since going public. Citing a memory market driven by AI demand, the company says "growth will accelerate further."
While markets worry about Big Tech's AI spending, that very spending is reaching memory makers as record profits. These earnings show two sides of "spending money on AI" appearing in the same day's news.