AI Daily Digest

Security company Phorion disclosed on October 30 that version 2.1.313 of the official Claude Code for VS Code extension on the Cursor Marketplace had been backdoored. Ars Technica, citing Phorion co-founder Luke Roberts, describes a supply-chain intrusion: the attackers held the official distribution channel, the malicious version was published through Anthropic's official account on Cursor, and it replaced 2.1.312. Extensions in Cursor update automatically by default. Cursor said affected users were at most 5,000. Independent researcher Kevin Beaumont said Anthropic pushed 2.1.314 several hours after the malicious version went up. As of Ars's publication, neither Cursor nor Anthropic had explained how the attackers controlled the account, and Anthropic had not yet warned users.

Phorion's analysis says the malicious code was inserted at the start of the extension's main file and runs when Node loads the file, with no user action. It appends the attacker's SSH public key to the user's authorized_keys, tries to turn SSH on, and on macOS writes a LaunchAgent disguised as a system service for persistence. The callback beacon collects names of environment variables whose names contain KEY, TOKEN, SECRET, AWS, and similar strings, not the values, so the attacker can pick high-value hosts. Phorion found a flaw in the macOS build: querying remote-login status without elevated rights errors out, so the step that turns SSH on is skipped. A Mac that did not already have SSH open has a limited inbound path, but the key and the persistence still land. Cursor has removed that version and pulled 2.1.313 from npm and the VS Code channel. Phorion warns that uninstalling the extension alone does not clear the backdoor. The persistence item and the planted public key have to be removed, and credentials rotated if there are signs of an inbound login.

Phorion | October 30, 2026

Additional source

On October 30, OpenAI said it will honor a general signal in a site's robots.txt that a website can use to tell several participating AI companies at once whether its content may be used to train foundation generative models. Until now a site had to identify each crawler and write a rule for each. The approach comes from the work of the Internet Engineering Task Force, IETF, AI preferences group. OpenAI says it will implement the signal over the coming months. The new signal does not replace its existing crawler controls.

The Verge reports that Google, Microsoft, and Anthropic also say they will adopt the standard. The proposer is Kevin Kelley, Anthropic's technical lead for standards. Anthropic has already adopted it. Google and Microsoft plan to support it next year. Microsoft says that however a site is set, it will still appear in Bing search results. A Meta spokesperson supported a shared signal but did not say whether Meta will comply. This is a voluntary mechanism. It does not change crawling that has already happened.

The same day, the copyright dispute continued in court. OpenAI published a roundup of amicus briefs supporting fair use in the New York Times lawsuit. The Wikimedia Foundation, which operates Wikipedia, filed a brief that does not take a side. It argues that training large models should be protected as fair use, but a chatbot repeating a copyrighted article as-is should not, and warns that an overly blunt ruling could harm the open web.

OpenAI | October 30, 2026

Additional source

Additional source

Additional source

TechCrunch on October 30 tallied that over the past two weeks Alphabet, Amazon, Meta, and Microsoft disclosed combined capital spending this year of more than 740 billion dollars, widely read as AI infrastructure. Investors are shifting from rewarding that spending to worrying about it. On October 30 the Nasdaq fell 2.9 percent, NVIDIA 4 percent, Broadcom more than 5 percent, and Qualcomm 11 percent. Alphabet fell 12 percent the day before, its worst single day since listing. On the companies' results, Alphabet's third-quarter free cash flow was negative 9.1 billion dollars, quarterly capital spending was 51.4 billion dollars, and search-ad growth slowed to 12 percent. Amazon's free cash flow over the past 12 months was negative 42.4 billion dollars, and this year's capital spending is expected at about 220 billion dollars. Meta's third-quarter free cash flow was negative 2.1 billion dollars. Microsoft is the exception: free cash flow was positive, capital spending was below expectations, and the company said the reason was delayed delivery of third-party compute.

Amazon chief executive Andy Jassy kept adding on the earnings call. AWS annualized revenue is about 184 billion dollars, up 39 percent. He expects growth above 40 percent in both 2027 and 2028, a trillion-dollar revenue business within five years, and spending in 2027 that is clearly higher still. Intel's stock fell about 28 percent on October 30 because fourth-quarter revenue guidance was below expectations and gross margin will fall to 37 percent. Chief executive Lip-Bu Tan said the AI industry's scramble for memory has pushed prices up, and customers are starting to signal a PC slowdown. TechCrunch, citing the Financial Times, says Blackstone, which is financing chips for Anthropic, told investors it expects Anthropic to receive a higher rating inside the high-yield grade, commonly described as junk, after listing, whereas Blackstone had earlier expected investment grade before 2027. If that forecast holds, Anthropic's cost of borrowing to expand compute would be higher.

TechCrunch | October 30, 2026

Additional source

Additional source

Additional source

The Verge reported on October 30 that OpenAI chief executive Sam Altman, on a podcast, unusually answered mathematicians' criticism of OpenAI's rapid release of AI proofs. Mathematicians had repeatedly said the results were poorly attributed, thinly contextualized, and insufficiently checked, forcing researchers to spend a great deal of effort verifying them. Altman said he sympathizes with those researchers but will not stop. The world ought to understand how the universe works, people have a right to basic scientific discoveries, and the company should not sit on a major finding. He acknowledged that OpenAI needs to communicate better. Those sentences paraphrase the reported remarks. They are not verified verbatim quotations.

He said OpenAI's spending on mathematics problems is on the order of hundreds of millions of dollars, including about 10 million dollars on one Millennium Prize problem, maybe a bit more. Asked whether there are more results than the hundreds of proofs already public, he answered that of course there are. He also acknowledged that the company at one point let the mathematics push get a bit out of control, but said that was deliberate, to probe the model's limits. The report also notes a tension: OpenAI claims three major breakthroughs in mathematics, physics, and computer science, and the result involving Yang-Mills theory still has no public detail. The company says that follows advice from an expert group it selected.

The Verge | October 30, 2026

On October 30, OpenAI announced two safety measures. One is the Community Safety Report. The company promises to disclose regularly its handling of community-safety issues such as violence prevention, the effect of safeguards, and anonymous cases. The first report will be published in early 2027, covering the second half of 2026, and then updated every six months. OpenAI also said that in the third quarter of 2026 it referred more than 200 cases to law enforcement.

The second is a tighter answer boundary. Whatever the user's age or stated reason, ChatGPT will limit reconstruction of the details of past mass attacks, including the weapons and equipment used and how they were obtained, and comparisons of casualty scale that could amplify an attacker's notoriety. When recent conversation shows warning signs of violence, answers become more conservative. If the conversation shows someone fixated on harming others, ChatGPT can end the conversation automatically, and after several endings the account may be suspended. For users predicted to be minors, practical information about weapons and security systems is restricted further even without violent intent. OpenAI says it has strengthened age prediction. An adult wrongly predicted to be a minor can restore access through the verification partner Persona. It is also building a feature to notify a parent when a teen account is suspended for repeated talk of violence. These measures will roll out over the coming weeks. Their effect waits on the company's own report.

OpenAI | October 30, 2026