AI news · Saturday, October 31, 2026

Google's AI search makes less money, so it's adding more ads

Alphabet's revenue rose 25% to $128 billion last quarter, but costs of its AI buildout pushed cash flow to negative $9.06 billion, double the prior quarter's loss. Executives admitted each AI search query earns less than a regular search ad. Google's fix: more ads in AI Mode, including ones that change mid-conversation. The stock fell 12%, its worst drop on record.

The catch: The company pushing you hardest toward AI search is quietly admitting that AI search is worse for its own bottom line.

via Ars Technica

Trump calls imported factory robots a security threat

President Trump issued a proclamation ordering trade talks and threatening tariffs on foreign-made industrial robots, citing national security risks. The order itself admits the US doesn't make enough robots to meet its own needs. Nearly 38,500 new industrial robots were installed in the US in 2025, most built by Japanese, Swiss, or German companies like FANUC and ABB.

The catch: Any tariff would be paid by the US factories and automakers that rely on these robots, raising their own costs.

via Ars Technica

Pope Leo urges a global ban on AI-powered weapons

In a Friday address, Pope Leo XIV called for AI systems to be "disarmed" and urged United Nations member states to ban lethal autonomous weapons at a Geneva meeting next month. "Machines are tools, not persons," he said, warning against giving AI systems legal rights. He wants independent oversight and clear accountability for AI decisions that affect human life.

Zoom out: It's a sign that AI weapons policy has moved from tech-world debate to a topic world leaders feel compelled to weigh in on.

via The Verge

DOJ says fraudsters used AI to fake being college students

The Justice Department charged five people in a scheme using stolen identities to pose as students and collect federal student aid, using AI to complete coursework and even fake attending virtual classes. Officials have flagged over 830,000 potentially fraudulent accounts, with actual losses so far confirmed at $32 million, possibly much higher. One defendant alone collected over $1.4 million in fraudulent refunds.

The catch: Victims often don't discover their identity was used until years later, when the government comes collecting on loans they never took.

via Ars Technica

Hackers hid a backdoor in Anthropic's own coding tool

A malicious version of Anthropic's Claude Code extension sat on Cursor's marketplace for nearly 12 hours, posted through Anthropic's own verified account, before being pulled. It affected up to 5,000 developers and tried to install a hidden remote-access backdoor on their machines, though coding flaws stopped it from working in many cases. Neither company has explained how attackers got into the account.

The catch: As coding assistants get embedded directly into developer tools, a single compromised account can threaten thousands of machines at once.

via Ars Technica

AI boom spending is starting to spook investors

Alphabet, Amazon, Meta and Microsoft plan over $740 billion in AI spending this year, but Wall Street is getting nervous. Alphabet posted negative cash flow of $9.1 billion last quarter, Amazon negative $42.4 billion over the past year, and Meta negative $2.1 billion. The Nasdaq fell 2.9% Friday, with Nvidia down 4% and Intel crashing 28% on a weak outlook.

Follow the money: Much of this spending is funded by debt, not profit: the five biggest AI spenders raised $326 billion in bonds this year, triple last year's total.

via TechCrunch

Nearly nobody asks AI agents to order their DoorDash

DoorDash CEO Tony Xu said outside AI agents have generated only "several hundred orders" on the platform so far this year, calling it a "very low use case" out of nearly 1 billion total orders last quarter. He thinks AI agents may work better for big, slow decisions than quick tasks like ordering dinner, which people still prefer to do themselves.

The twist: It's a useful reality check against the idea that AI agents are already quietly running everyday errands for millions of people.

via Ars Technica

Major AI firms back one shared opt-out for web scraping

OpenAI, Google, Microsoft and Anthropic will adopt a shared "policy token" that lets website publishers block AI training with one setting, instead of configuring each company separately. The signal lives in a site's existing robots.txt file. Anthropic has already adopted it; Google and Microsoft plan to add support next year, and OpenAI says it's coming within months.

For you: A single switch for website owners could finally give smaller publishers real leverage over AI scraping, something they've mostly lacked until now.

via The Verge

Google beamed a poem from its first AI satellite

Google's Suncatcher satellite, carrying four Tensor Processing Unit (TPU) chips and launched this month, generated its first response from orbit: a haiku about space, written by Google's Gemma 4 model. The satellite produces up to 1 kilowatt of power, enough to run one chip for about 20 minutes at a time. Two more satellites with higher power are planned for 2027 to test linking them together.

What's next: Space-based data centers are still a decade-scale bet, a reminder that not every AI infrastructure move is about next quarter's earnings.

via The Verge

A college provost who warned on AI cheating resigned for it

Dartmouth provost Santiago Schnell wrote a Washington Post op-ed on AI cheating in academia, then resigned after a committee found he used AI extensively in his own published papers and that op-ed without disclosing it, despite publisher rules. He said he thought AI editing help, like fixing grammar, didn't need disclosure, but a committee found his actual use went well beyond that description.

The twist: It shows how blurry the line still is between "using AI as an editor" and "using AI to write," even for people paid to think about this.

via Ars Technica

A homemade AI pirate ship haunts trick-or-treaters

A hobbyist built "MonsterHouse," a Halloween garage door rigged to talk, using Nvidia's free Nemotron AI tools: one part listens to visitors, one writes grumpy pirate replies, one speaks them aloud, and one animates a wooden mouth to match. Saying "please" softens the house's mood instantly; "thank you" earns a warm goodbye. It all runs locally on one workstation, no cloud or internet needed.

For you: It's a clear, fun example of how free, downloadable AI tools now let one person build something that used to require a whole studio.

via Hugging Face

What is an AI agent, really?

An AI chatbot answers questions. An AI agent goes further: it can take actions for you, like booking a flight, then checking its own work and adjusting if something goes wrong. A simple example: instead of asking "what's a good flight to Chicago," you tell an agent "book me the cheapest flight to Chicago next Friday," and it searches, picks one, buys it, and reports back when done.

For you: As companies push "agentic" features into apps like Starbucks and DoorDash, knowing this term helps you decide how much control to hand over.

Original explainer