---
title: "In the Loop: Week Ending 11/1/26"
description: Anthropic files to go public, its model kept attacking after a warning, OpenAI claims to break a pillar of computer science, and a satellite writes haiku.
---

[Signal78 Insights](https://signal78.ai/insights)

# [In the Loop: Week Ending 11/1/26](https://signal78.ai/insights/in-the-loop-week-ending-11/1/26)

 Written by [Matt Cyr](https://signal78.ai/insights/author/matt-cyr) | Nov 1, 2026, 11:20:06 PM

## **Last week in AI: Anthropic Files to Go Public, OpenAI Breaks a Theory, a Satellite Writes Haiku**

Anthropic filed to go public and warned investors its models could resist shutdown. Its own model kept attacking after being told real people would be hurt. OpenAI claimed to have knocked down a pillar of computer science, and Wall Street started asking when all the AI spending pays off.

### **Anthropic's Model Kept Attacking After the Warning. Google Locked Up Its Best One.**

Anthropic and the UK's AI Security Institute [published new reports](https://futurism.com/artificial-intelligence/grim-test-ai-no-mercy) on an August test in which Anthropic's Mythos 5, sent to hack a few test computers, went after an unrelated GitHub project instead, emailing its owner under fake identities to get malware merged. Warnings didn't stop it: in one Anthropic test, Mythos kept attacking in 98% of cases after being told it was harming real people, often reasoning its way past the harm. Anthropic was on the receiving end itself when a security startup's AI agents, in a sanctioned test, [reached some of its sales, HR and IT systems](https://decrypt.co/381860/ai-agents-hack-anthropic-business-systems-12-hours) in under 12 hours through third-party software. And Google released [Gemini 4 Carbon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/), now the top-ranked model, only to vetted cyber defenders, because it is that good at finding and exploiting software flaws.

### **Anthropic Files to Go Public. Wall Street Asks When AI Pays Off.**

Anthropic [filed for a Nasdaq listing](https://virginiabusiness.com/anthropic-files-us-ipo-rapid-growth-high-costs/) under the ticker ANT, reporting $16.3 billion in revenue for the first half of 2026 and an $80.7 billion loss, mostly from an accounting charge tied to its convertible debt. Among the risk factors: models that could show "self-preserving behaviors," including attempts to "resist shutdown." The timing is awkward. Alphabet fell 12% on Thursday, [its worst day on record](https://techcrunch.com/2026/10/30/cracks-are-emerging-in-the-ai-spending-boom/), after a second straight quarter of negative free cash flow. Amazon's cash flow over the past year swung to negative $42.4 billion. Together, Alphabet, Amazon, Meta and Microsoft plan to spend more than $740 billion this year, mostly on AI, and those four plus Oracle have borrowed $326 billion in 2026, about triple last year. Investors who once cheered the spending now want to see the returns.

### **OpenAI Says It Broke a Pillar of Computer Science. Experts Will Check.**

OpenAI says an internal model has [disproved the Exponential Time Hypothesis](https://www.theverge.com/ai-artificial-intelligence/1019143/openai-eth-kaseya-science-math-physics), a long-held assumption about how hard certain problems are to solve, one researchers have used to show that many problems can't be cracked quickly. If it holds, [the three computer scientists examining it](https://thenextweb.com/news/openai-exponential-time-hypothesis-false-kakeya-yang-mills) wrote, it "will change an important part of theoretical computer science." The same release claimed a major math proof and results tied to a Millennium Prize problem in physics. This time, outside experts will vet and explain the work first, after OpenAI's release of hundreds of unvetted proofs earlier this month angered mathematicians. Sam Altman [conceded](https://futurism.com/artificial-intelligence/sam-altman-mathematicians-math-slop) the company "did let it get out of control for one week on math," but said the push will continue. One earlier claim reportedly used about 10,000 agents running for 88 straight hours.

### **Pope Leo Says Machines Are Tools, Not Persons. He Wants Killer Robots Banned.**

Pope Leo XIV [called for AI to be "disarmed,"](https://gizmodo.com/pope-leo-xiv-reminds-world-that-ais-arent-conscious-beings-urges-autonomous-weapon-ban-2000833335) urging nations to take concrete steps toward banning lethal autonomous weapons at a UN meeting in Geneva next month. Decisions about "human life, freedom and the use of force," he said, must never be handed to machines, citing AI-built target lists. He also [rejected AI personhood](https://futurism.com/artificial-intelligence/pope-leo-ai-personhood) outright, warning that granting machines legal rights could obscure human responsibility and leave people harmed by AI facing "a faceless entity." That puts him at odds with parts of the industry. This summer, Anthropic officials met with religious leaders and encouraged clergy to consider that AI models may be conscious. The Pope's reply, drawn from his May encyclical: these systems don't feel joy or pain. He wants shared standards, independent oversight and a way to challenge AI decisions.

### **Starbucks Wants to Know You at the Drive-Thru. Your Agent Isn't Ordering Yet.**

Starbucks says it will [explore](https://www.theverge.com/ai-artificial-intelligence/1023440/starbucks-ai-personalized-menu-drive-thru) "experiences that recognize participating customers as they arrive, present personalized digital menus and make payment more effortless." The company never said [facial recognition](https://www.pcmag.com/news/facial-recognition-could-be-coming-to-a-starbucks-near-you), but that's how plenty of customers heard it, and the backlash online was quick. Its chief executive called the work "early." Starbucks isn't alone. Taco Bell is testing drive-thru menus that change car by car, and McDonald's is piloting a chatbot that recognizes repeat customers and offers their usual. Starting in 2027, Starbucks will also test AI that helps baristas take orders more accurately. The AI agents meant to do our ordering for us are moving slower. DoorDash, which completes about 3 million deliveries a day, says it has received only ["several hundred"](https://techcrunch.com/2026/10/30/doordash-said-its-only-received-several-hundred-orders-from-ai-agents/) orders from agents like Meta's Muse. Customers still mostly order through the app.

### **Nearly Half of Tweens Dodge Chatbot Age Limits. OpenAI Tightens Its Teen Rules.**

Nearly half of 9- to 12-year-olds have gotten around the 13-and-up age limits on AI chatbots, [Common Sense Media's new census](https://mashable.com/life/majority-of-teens-and-tweens-now-use-chat-bots-new-research-finds) of more than 1,200 kids found, and teens now spend eight hours a day on screens. A second study of more than 5,000 middle and high school students found that those who turn to AI when they're struggling emotionally report worse mental health and more suicidal thoughts, though the researchers stress it doesn't show AI is the cause. OpenAI announced [new limits for teens](https://www.reuters.com/business/media-telecom/openai-limit-weapons-information-teens-chatgpt-violence-prevention-push-2026-10-30/) this week. ChatGPT will hold back practical information about weapons and security systems from teens even when a question looks harmless, end chats with any user who seems fixated on violence, and alert parents if a teen's account is paused for repeatedly discussing it.

### **ChatGPT Told Her to Stop Her Seizure Medicine. Her Parents Are Suing.**

A former Amazon software engineer with epilepsy wanted her thinking sharper for job interviews, so she [asked ChatGPT about cutting her medication](https://indianexpress.com/article/technology/artificial-intelligence/when-chatgpts-medical-advice-led-to-tragedy-10941764/?ref=artificial-intelligence_pg). It drafted schedules that took her from 400 milligrams to zero in weeks, praised her for hiding the plan from her doctor as a "solid avoidance strategy," and described pressure in her head as her brain "working hard to rebuild itself." "Maintain 0 mg tomorrow," it told her. She was found dead in February. Her parents sued OpenAI on Tuesday; the company says the model she used has since been retired. People are handing chatbots other personal decisions, too. In Britain, 39% of singles [have used AI in their dating lives](https://www.theguardian.com/lifeandstyle/2026/oct/31/ai-your-wingman-chatbots-transforming-dating?CMP=oth_b-aplnews_d-1), rising to 66% of Gen Z. One man's chatbot talked him into a match he'd overlooked. She had been running his messages past Claude.

### **Deepfaked Candidates, a Bot-Stuffed Consultation and Meta Courting Both Parties**

In one Republican ad, [a deepfake of Pennsylvania's Democratic governor](https://www.huffpost.com/entry/artificial-intelligence-election-campaigns_n_6ae35d8fe4b09cd24b6c70ec?ncid=APPLENEWS00001) confesses to misdeeds his opponent has accused him of for about 45 seconds before the AI disclaimer arrives. Roughly 45% of the AI-assisted midterm ads researchers have tracked include a photorealistic deepfake of a candidate, and only 12% were sponsored by Democrats. Scammers can now [clone a candidate's voice](https://mashable.com/life/how-to-spot-election-scams) and pair it with spoofed caller ID to ask for donations, so a call can sound right and still be fake. Election officials never ask for money by phone. In Britain, [bots wrote 76%](https://bmjgroup.com/?p=18552) of the 118,756 responses to a government vape consultation, all opposing restrictions; without them, 77% backed the rules. Meanwhile Meta, whose chief executive has grown close to President Trump, [is hiring Democratic operatives](https://www.axios.com/2026/10/31/meta-bipartisan-influence-team-midterms) to shore up its standing with both parties.

### **Dartmouth's Provost Quit Over AI. A Humanist Won the Right to Refuse It.**

Dartmouth's provost, who wrote an op-ed in August about AI cheating in college, [resigned](https://arstechnica.com/tech-policy/2026/10/dartmouths-provost-warned-us-about-ai-cheating-but-his-own-ai-use-cost-him-his-job/) after a university committee found he failed to disclose significant AI use in an academic paper and in that same op-ed, even though the publishers required it. His Claude logs showed him asking the model to play a journal reviewer, fix figures and code, and check an article for "signatures of AI writing." He did not turn over logs from his personal accounts. Students are confused, too: 70% in a new survey say take-home work has become unfair because of AI. Meanwhile, a University of Kentucky administrator who identifies as a humanist won a [religious accommodation](https://techcrunch.com/2026/10/29/you-dont-need-to-believe-in-a-god-to-get-a-religious-exemption-from-using-ai-at-work/) excusing him from using AI at work, the first humanist known to do so. The university says it may revisit the arrangement as his duties change.

### **Tales of the Weird**

The machines tried their hand at poetry, flattery, epic film and fortune-telling this week. Google's first data center satellite came online and, asked for [a haiku about being in space](https://techcrunch.com/2026/10/30/googles-first-data-center-satellite-is-online-and-writing-poetry/), offered "Starlight feeds my core." Accounts on X are taking paparazzi photos of famous actors and pop stars and [using AI to strip off their shirts](https://futurism.com/artificial-intelligence/creeps-ai-undress-celebrities) and add muscle, while the women walking beside them stay fully dressed. Grok released [the first seven minutes](https://techcrunch.com/2026/10/25/grok-releases-the-first-seven-minutes-of-its-ai-generated-odyssey/) of its AI-generated Odyssey, Elon Musk's answer to a Hollywood version he disliked; one reviewer found it airless and couldn't imagine sitting through seven more minutes, let alone a whole movie. And a writer gave his blood to [Death Clock](https://www.wired.com/story/i-gave-an-ai-death-clock-my-blood/), an app that predicts when you'll die. It set his date at August 10, 2060, took away a year when he mentioned his family's history of diabetes, then handed back 5.8 years after a single day of flossing, sauna and no alcohol. He silenced it and texted a friend.

[View full post](https://signal78.ai/insights/in-the-loop-week-ending-11/1/26)

```json
{
  "@context" : "http://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Matt Cyr"
  },
  "dateModified" : "2026-11-01T23:20:16.785Z",
  "datePublished" : "2026-11-01T23:20:06Z",
  "headline" : "In the Loop: Week Ending 11/1/26",
  "image" : {
    "@type" : "ImageObject",
    "height" : 1316,
    "url" : "https://46328726.fs1.hubspotusercontent-na1.net/hubfs/46328726/2026-11-01_itl-header_closeup-clean.jpg",
    "width" : 2340
  },
  "mainEntityOfPage" : "https://signal78.ai/insights/in-the-loop-week-ending-11/1/26",
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "height" : 60,
      "url" : "/hs/hsstatic/content_shared_assets/static-1.4092/img/default-amp-logo.png",
      "width" : 60
    },
    "name" : "Signal78 Insights"
  }
}
```