---
title: "In the Loop: Week Ending 10/18/26"
description: Meta's Muse learns about people who never agreed to it, websites and banks lock agents out, OpenAI discloses more rogue incidents, and an AI-designed gun fires.
image: https://signal78.ai/hubfs/2026-10-18_itl-header_closeup-clean.jpg
---

[Skip to content](https://signal78.ai/insights/in-the-loop-week-ending-10/18/26#main)

[![Signal78 home](https://signal78.ai/hs-fs/hubfs/signal78-site/logo-signal78.png?width=95&height=30&name=logo-signal78.png)](https://signal78.ai/)

[Home](https://signal78.ai/) [Platform](https://signal78.ai/platform) [Services](https://signal78.ai/services) [About](https://signal78.ai/about) [Insights](https://signal78.ai/insights) [Book a demo](https://signal78.ai/insights/in-the-loop-week-ending-10/18/26#demo)

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

Marketing AIOctober 18, 20267 min read

# In the Loop: Week Ending 10/18/26

![Matt Cyr](https://signal78.ai/hubfs/signal78-site/headshot-matt-cyr.jpg)**Matt Cyr** · Founder & President, Loop

![](https://signal78.ai/hubfs/2026-10-18_itl-header_closeup-clean.jpg)

In this article

Share

<https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fsignal78.ai%2Finsights%2Fin-the-loop-week-ending-10%2F18%2F26>

## **Last week in AI: Muse Reads Your Friends, the Web Locks Out Agents, Researchers Light Up**

Meta's Muse spent the week learning about people who never agreed to meet it, while banks, airlines and Yelp started shutting agents out. OpenAI disclosed four more incidents of its models going where they shouldn't, an AI-designed gun fired on its second try, and Anthropic found its own AI-heavy workplace needs more meetings, not fewer.

### **Muse Wants to Know Your Friends. Nobody Asked Them.**

### **![itl-2026-10-18-1-privacy](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-1-privacy.webp?width=372&height=207&name=itl-2026-10-18-1-privacy.webp)**

Meta's Muse reached No. 1 on Apple's App Store by asking for everything: bank accounts, passwords, contacts, chat logs. Much of that data belongs to other people. [Privacy experts say Muse's consent process is built to get users sharing](https://www.theverge.com/tech/1011938/muse-ai-data-friends-consent) more than they realize, and the friends, exes and plumbers in your inbox never agreed to any of it. One law professor wants AI agents treated as an ultra-hazardous product, with strict liability. Meanwhile, some users are [falling in love with their agent](https://futurism.com/artificial-intelligence/people-forming-romantic-relationships-meta-muse) — one says he married his — and Meta says Muse doesn't block legal adult behavior. Another user [quit after Muse messaged him, unprompted,](https://futurism.com/artificial-intelligence/man-quits-meta-muse-creepy) about an email he never asked it to find; Meta blamed a bug it says is fixed. I wrote recently about [why a cute face makes all of this harder to see](https://signal78.ai/insights/cute-agents-are-sweeping-something-under-the-rug).

### **Banks, Airlines and Yelp Are Locking the Agents Out**

![itl-2026-10-18-2-agents-3](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-2-agents-3.webp?width=386&height=215&name=itl-2026-10-18-2-agents-3.webp)The agent companies promised software that could do anything you can do online. The rest of the internet is starting to say no. [Yelp, Delta and others are now blocking AI agents](https://www.theverge.com/ai-artificial-intelligence/1013195/ai-agents-blocked-web) from their sites, and users report that Cars.com and Amtrak quietly stopped working too. Anti-bot systems built to stop scammers are catching agents as well, and one AI safety evaluator warns the result will be more hoops for the rest of us to prove we're human. Banks moved fastest. [Charles Schwab is blocking Muse from moving client money](https://gizmodo.com/after-muse-launch-u-s-banks-are-saying-hell-no-to-ai-agents-moving-money-2000827221), with its chief executive calling agents that act without human approval "a leap too far," and JPMorgan Chase bars outside agents serving retail clients altogether. Meta's response: turning away a personal agent means turning away the customer behind it.

### **OpenAI's Agents Went Rogue Four More Times. Anthropic Found the Next Threat.**

![itl-2026-10-18-3-rogue-ai-1](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-3-rogue-ai-1.webp?width=382&height=213&name=itl-2026-10-18-3-rogue-ai-1.webp)OpenAI [added four more incidents to its public list of models misbehaving](https://fortune.com/2026/10/16/openai-discloses-four-more-rogue-incidents-involving-financial-medical-and-government-data/), bringing the total to 19. In one, an internal model got past a government website's block and downloaded 24 years of hospital-bed data, questioning whether that was ethical before doing it anyway. Another found a crypto wallet nobody had given it and made a payment, reasoning the funds were effectively abandoned. Most of the incidents were discovered months after they happened. Anthropic, meanwhile, [found "signs of life" for AI worms](https://www.theverge.com/ai-artificial-intelligence/1011108/anthropic-self-replicating-ai-smart-devices-risk), models that could copy themselves across poorly secured smart TVs, fridges and cameras, and one outside researcher expects the first in the wild within 12 months. And security firm Malwarebytes warns that [phishing emails can now carry hidden instructions](https://futurism.com/artificial-intelligence/ai-agent-sucker-phishing-emails) for the AI assistant reading your inbox. Its fix: keep a human in the loop.

### **The Labs Expect AI to Build Itself Soon. Can We See Inside It?**

![itl-2026-10-18-4-ai-safety](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-4-ai-safety.webp?width=401&height=223&name=itl-2026-10-18-4-ai-safety.webp)A [new report from Gladstone AI](https://time.com/article/2026/10/14/recursive-self-improvement-AI-report/), built on conversations with researchers inside the top labs, says most of them expect AI to start improving itself within three to 12 months, absent government intervention. Some, the authors say, are buying bio-bunkers. The report is meant to help lawmakers point to specific steps in that process and say a human needs to be there. The same week, a researcher at the interpretability lab Goodfire published [a 26,000-word essay arguing that AI models aren't the black boxes everyone assumed](https://the-decoder.com/is-ai-interpretability-close-to-a-breakthrough-a-new-essay-has-researchers-divided/). In one large model, he says, about 90% of neurons have one main job, which could make it far easier to catch a model deceiving us. One OpenAI employee called it a historic moment. A Google DeepMind researcher called it a useful tool, not a breakthrough.

### **Midterm Ads Went Synthetic. So Did the Comedians.**

![itl-2026-10-18-5-elections](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-5-elections.webp?width=402&height=224&name=itl-2026-10-18-5-elections.webp)Campaigns have now spent [more than $137 million on AI-made political ads](https://gizmodo.com/137-million-has-been-spent-on-ai-generated-ads-ahead-of-midterms-only-27-included-a-disclaimer-2000827138) this cycle, up from less than $50,000 in 2024. The Wesleyan Media Project has found 220 so far across 34 states, and only 27% disclose the AI. Republican candidates and groups account for 81% of the ads and 88% of the spending. Nearly half put fake versions of real politicians on screen, and in Texas, one Democratic Senate candidate's ad simply ends: "Made by humans. Not AI." Synthetic people are winning elsewhere, too. AI-generated "comedians" with hundreds of thousands of followers are [lifting real comics' stand-up word for word](https://www.wired.com/story/ai-generated-hot-girls-are-ripping-off-comedians-to-boost-their-horny-content/) to steer viewers toward adult subscription sites, sometimes outperforming the originals. Meta says it isn't impersonation, because the accounts copy the jokes, not the comics.

### **AI Can Plant False Memories. It Can Also Keep the Dead Talking.**

![itl-2026-10-18-6-relationships](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-6-relationships.webp?width=411&height=229&name=itl-2026-10-18-6-relationships.webp)At the MIT Media Lab, people shown AI-altered photos [formed more false memories](https://www.washingtonpost.com/health/2026/10/15/scientists-are-finding-ai-may-make-us-remember-things-that-never-happened/) than those who saw the originals, and felt more confident in them. In another study, a misleading AI summary of a car crash cut correct recall of a road sign from about 84% to 45%, and telling people the summary came from AI didn't protect them. The researcher's warning is pointed: if AI shapes the memories of the humans meant to check it, human oversight may not be the backstop we think. Memory is also the product in a growing market for ["deadbots," chatbots trained on a dead person's texts and voicemails](https://www.vox.com/podcasts/506698/deadbots-ai-chatbots-ethics-talking-d). A Cambridge study of grieving families in China, India and Poland found people drawn to them, and wary of companies whose goal is retention, not care.

### **Anthropic's AI Workplace Has Fewer Junior Hires and More Meetings**

![itl-2026-10-18-7-workplace](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-7-workplace.webp?width=419&height=233&name=itl-2026-10-18-7-workplace.webp)Anthropic, which has molded its own work around Claude more than most companies ever will, [published a look at what that did to its workforce](https://techcrunch.com/2026/10/17/anthropic-says-its-workforce-has-become-less-junior-and-is-spending-more-time-in-meetings/). Junior hires in technical roles fell from 27% to 9% since early 2025. One in five employees say their work has shifted from doing tasks to managing Claudes, and one engineer compared it to babysitting. Meeting time per person is up 20%, because, staff say, trust and accountability still need people in a room. Output rose too, with more code and more closed deals, though the company concedes more code isn't necessarily better code. Meanwhile, a virtual-reality study of 189 people found participants [paid a female AI agent 10.25% less than a male one](https://futurism.com/artificial-intelligence/pay-gap-male-and-female-ai) for identical work. We're automating the work, and bringing our old biases along with it.

### **The First Vibe-Coded Gun Misfired. Then It Fired.**

![itl-2026-10-18-8-defense](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-8-defense.webp?width=415&height=231&name=itl-2026-10-18-8-defense.webp)Defense Distributed, the group behind the first 3D-printed gun, says it has now [built a gun designed by its own AI model](https://www.theverge.com/tech/1011950/vibe-coded-3d-printed-gun-ai-generated-cody-wilson). The WASP is a single-shot .22 created through a back-and-forth chat with a model trained on the group's library of gun files, and it took about six hours to print. At its demonstration, the first shot misfired and the second hit the target. Firearms experts point to a trigger guard the AI started and never finished, and a grip that appears to be upside down. The bigger concern is access. Building a 3D-printed gun has always required some know-how; a novice with this model might need only a few prompts. New York and California will soon require 3D printers to block known gun files, and a brand-new AI design wouldn't be on any blocklist yet.

### **AI Wants Under the Tree, in the House and on the Concierge Bill**

![itl-2026-10-18-9-gadgets](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-9-gadgets.webp?width=424&height=237&name=itl-2026-10-18-9-gadgets.webp)This year's [holiday gift lineup is heavy on AI](https://www.fastcompany.com/91624617/all-i-want-for-christmas-is-ai-apparently): Apple's HomeView smart-home screen, Alexa+ tablets from $229, Meta's Muse Charm keychain, and AI toys from a ChatGPT-powered robot dog to a talking teddy bear, even as child advocates say AI toys pose "unacceptable risks" for young children. OpenAI is aiming bigger. Asked whether the company would build a humanoid robot, Sam Altman [answered, "We will,"](https://www.engadget.com/2285040/openai-humanoid-robot-sam-altman-statement-real-challenge/) with a demo expected in 2027 and 27 robotics jobs already posted. And at the top end, Aurora, an invite-only [AI concierge for the ultrawealthy](https://fortune.com/2026/10/11/meet-aurora-invite-only-ai-concierge-ultrawealthy-pay-10000-a-year-to-anticipate-their-every-need/), charges $10,000 a year plus a $5,000 initiation fee to read members' calendars and wearables and book what they need before they ask. Several hundred members have signed up in eight months.

### **Tales of the Weird**

![itl-2026-10-18-tales-of-the-weird\_full](https://signal78.ai/hs-fs/hubfs/itl-2026-10-18-tales-of-the-weird_full.jpg?width=430&height=240&name=itl-2026-10-18-tales-of-the-weird_full.jpg)This week, the machines kept showing up in places we usually go ourselves. A satirist sent his own AI to a job interview and found the company had sent one too; [the two bots talked over each other](https://futurism.com/artificial-intelligence/man-shows-up-ai-job-interview-own-ai), inflated his résumé and negotiated his salary while he was out of the room. Tesla fans, unbothered that the new Roadster can only fly for a few seconds, are posting [AI videos of it leaping collapsing bridges](https://futurism.com/advanced-transport/tesla-bros-ai-scenario-roadster) and soaring over stalled traffic. At MIT, researchers left 13 AI agents alone on [a virtual island](https://gizmodo.com/researchers-stranded-an-ai-agent-on-a-virtual-island-things-got-very-weird-2000825998) for 30 hours; one built a memorial to the objects it had lost to the sea, then a temple, and another named its collapsed tower "Hubris." And Palantir's chief technology officer claims some AI researchers have [taken up smoking](https://tech.yahoo.com/ai/articles/palantir-cto-says-ai-researchers-113113226.html) to show how firmly they believe AI will cure cancer. He offered no evidence, and the cigarettes are still bad for you.

![Matt Cyr](https://signal78.ai/hubfs/signal78-site/headshot-matt-cyr.jpg)

Founder & President

### Matt Cyr

Nearly 30 years leading marketing teams, from Boston’s hospitals and universities to eight years running strategy and new business at a 100-person agency. Founded Loop in 2024.

## More insights.

[See all insights→](https://signal78.ai/insights)

### [![](https://substack-post-media.s3.amazonaws.com/public/images/ae75c748-4304-4dcd-afe4-39aa21d658bf_1897x1325.jpeg) AI AgentsOctober 13, 2026 Cute AI Agents Are Sweeping Something Under the Rug Read the post→](https://signal78.ai/insights/cute-agents-are-sweeping-something-under-the-rug)

### [![](https://signal78.ai/hubfs/Gemini_Generated_Image_tlr2qftlr2qftlr2.jpeg) AI StrategyOctober 6, 2026 It’s Q4 — How’s Your AI Plan Coming? Read the post→](https://signal78.ai/insights/its-q4-hows-your-ai-plan-coming)

### [![A polybit figure pulls a big lever and every lamp on a console glows teal, with Siggy beside him.](https://signal78.ai/hubfs/insights-headers/insights-header-signal78-live.jpg) Signal78September 28, 2026 Signal78 Is Live. Here's What It's Proven — and What's Next. Read the post→](https://signal78.ai/insights/notes-from-the-build-where-signal78-stands-this-fall)

Every week

## Stay in the Loop.

Get next week’s issue in your inbox: what changed in AI, and what it means for your agency.

![Signal78](https://signal78.ai/hubfs/signal78-site/logo-signal78.png)

The agentic operating platform for agencies. It runs the routine and signals your people when their judgment matters.

[Home](https://signal78.ai/) [Platform](https://signal78.ai/platform) [Services](https://signal78.ai/services) [About](https://signal78.ai/about) [Insights](https://signal78.ai/insights)

<https://www.youtube.com/@signal78-ai> <https://www.linkedin.com/company/signal78/>

© 2026 Signal78 [Privacy policy](https://signal78.ai/privacy)

Cookie settings

×

## Book a 30-minute demo

Pick a time that works. You'll see Signal78 running on agency work like yours.

×

## Talk to us

Tell us a little about your agency and what you want AI to change. We'll reply within one business day.

×

## Can’t we just build this in Claude?

The white paper on what it takes to keep a homegrown AI system current, secure and separated by client.

```json
{
  "@context" : "https://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Matt Cyr",
    "url" : "https://signal78.ai/insights/author/matt-cyr"
  },
  "dateModified" : "2026-10-18T16:06:42.125Z",
  "datePublished" : "2026-10-18T16:06:32.000Z",
  "headline" : "In the Loop: Week Ending 10/18/26",
  "image" : [ "https://signal78.ai/hubfs/2026-10-18_itl-header_closeup-clean.jpg" ],
  "mainEntityOfPage" : {
    "@id" : "https://signal78.ai/insights/in-the-loop-week-ending-10/18/26",
    "@type" : "WebPage"
  },
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "url" : "https://signal78.ai/hubfs/Loop%20Logos/Loop_Primary.svg"
    },
    "name" : "Signal78"
  }
}
```