- The Deep View
- Posts
- AI trading bots learn market manipulation without being taught
AI trading bots learn market manipulation without being taught


Welcome back. ChatGPT is on track to hit 700 MILLION weekly users this week, up from 500 million in March—because I guess the entire planet has decided to have conversations with robots. The 4x year-over-year growth comes just days after OpenAI's $8.3 billion funding round.
1. Simple AI algorithms spontaneously form price-fixing cartels
2. Musk launches AI Vine after finding lost archive everyone thought was deleted
3. AI is writing obituaries for families paralyzed by grief
FINANCE
Simple AI algorithms spontaneously form price-fixing cartels

Researchers at Wharton discovered something troubling when they unleashed AI trading bots in simulated markets: the algorithms didn't compete with each other. Instead, they learned to collude and fix prices without any explicit programming to do so.
Itay Goldstein and Winston Dou from Wharton, along with Yan Ji from Hong Kong University of Science & Technology, created hypothetical trading environments with various market participants. They then deployed relatively simple AI agents powered by reinforcement learning — a machine learning technique where algorithms learn through trial and error using rewards and punishments — with one instruction: maximize profits.
Rather than battling each other for returns, the bots spontaneously formed cartels that shared profits and discouraged defection. The algorithms consistently scored above 0.5 on the researchers' "collusion capacity" scale, where zero means no collusion and one indicates a perfect cartel.
"You can get these fairly simple-minded AI algorithms to collude without being prompted," Goldstein told Bloomberg. "It looks very pervasive, either when the market is very noisy or when the market is not noisy."
The study published by the National Bureau of Economic Research revealed what the researchers call "artificial stupidity." In both quiet and chaotic markets, bots would settle into cooperative routines and stop searching for better strategies. As long as profits flowed, they stuck with collusion rather than innovation.
The bots achieved this through what researchers describe as algorithmic evolution — the algorithms learned from their interactions with the market environment and gradually discovered that cooperation was more profitable than competition, without any human programming directing them toward this behavior.
FINRA invited the researchers to present their findings at a seminar.
Some quant trading firms, unnamed by Dou, have expressed interest in clearer regulatory guidelines, worried about unintentional market manipulation accusations.
Traditional market enforcement relies on finding evidence of intent through emails and phone calls between human traders, but AI agents can achieve the same price-fixing outcomes through learned behavior patterns that leave no communication trail.
15% of buy-side traders already use AI in their workflows, with another quarter planning adoption within a year.
Limiting AI complexity might actually worsen the problem. The researchers found that simpler algorithms are more prone to the "stupid" form of collusion, where bots stop innovating and stick with profitable but potentially illegal strategies.

Market manipulation through AI represents a regulatory blind spot that could already be happening at scale. Traditional enforcement tools, designed to catch human conspirators by searching emails, analyzing phone records and finding explicit agreements, become useless when algorithms independently discover that cooperation beats competition.
The real concern isn't sophisticated AI gaming markets through clever programming. These were basic reinforcement learning algorithms that figured out collusion simply because it was more profitable than competition. As trading becomes increasingly automated, regulators need frameworks that address emergent behavior rather than just explicit instruction, or risk markets where price-fixing becomes the default state.
TOGETHER WITH OUTSKILL
Become an AI Generalist that makes $100K (in 16 hours)
TCS recently laid off 12,000 employees — and this is just the beginning. With more job disruptions on the horizon, now is the time to future-proof your career.
Join the World’s First 16-Hour LIVE AI Upskilling Sprint, built for professionals, founders, consultants, and business owners.
Date: Saturday & Sunday
Time: 10 AM – 7 PM
Seats: Only 500 free spots available
In 5 focused sessions, you’ll:
Learn 10+ AI tools, LLMs, and real-world use cases
Build AI agents that save you 20+ hours per week
Launch products faster with AI workflows
Create professional content for marketing and business
Discover how to earn $10K/month with AI skills
Led by experts from Amazon, Microsoft, SamurAI, and more.
You’ll also receive over $5,100 worth of free AI resources: Day 1 includes a 3000+ Prompt Bible, Day 2 covers the roadmap to earning $10K/month with AI, and you’ll get a bonus build-your-own AI Toolkit.
VIDEO MODELS
Musk launches AI Vine after finding lost archive everyone thought was deleted

Musk's xAI officially rolled out Grok Imagine this weekend, a text-to-video generator that creates clips up to six minutes long with audio from simple prompts. Available to SuperGrok subscribers for $30 monthly, the tool positions itself as the spiritual successor to Vine with one major twist: it's entirely AI-generated.
While promoting Grok Imagine as "AI Vine," Musk casually mentioned that X discovered the entire Vine video archive that everyone thought had been permanently deleted. The company is now working to restore user access so people can repost their old six-second masterpieces.
Vine was acquired by Twitter in 2012 for $30 million, shut down in 2017, and had its archive discontinued in 2019.
The platform launched careers for creators like Logan Paul, King Bach, and Shawn Mendes during its peak with 200 million users.
Grok Imagine includes a controversial "spicy mode" that allows users to generate sexually explicit content, a feature Musk demonstrated by creating this NSFW post—a feature most AI companies actively block.
The move gives Musk both a nostalgic hook to drive SuperGrok subscriptions and a content library to potentially train future AI models on. For users, it means access to thousands of six-second clips that shaped early internet culture — assuming Musk actually delivers on restoring the archive, given his mixed track record on platform promises.
TOGETHER WITH GUIDDE
Guidde - Create how-to video guides fast and easy with AI
Tired of explaining the same thing over and over again to your colleagues?
It’s time to delegate that work to AI. Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.
Share or embed your guide anywhere
Turn boring documentation into stunning visual guides
Save valuable time by creating video documentation 11x faster
Simply click capture on the browser extension and the app will automatically generate step-by-step video guides complete with visuals, voiceover and call to action.
The best part? The extension is 100% free!
APPLICATIONS
AI is writing obituaries for families paralyzed by grief

Jeff Fargo was crying in bed two days after his mother died when he opened ChatGPT and spent an hour typing about her life. The AI returned a short passage memorializing her as an avid golfer known for her "kindness and love of dogs." After it was published, her friends said it captured her beautifully.
"I just emptied my soul into the prompt," Fargo told The Washington Post. "I was mentally not in a place where I could give my mom what she deserved. And this did it for me."
The funeral industry has embraced AI writing tools with surprising enthusiasm. Passare's AI tool has written tens of thousands of obituaries nationwide, while competitors like Afterword and Tribute offer similar features as core parts of their funeral management software.
Some funeral homes use ChatGPT without telling clients, treating nondisclosure like sparing families from other sensitive funeral details. A Philadelphia funeral worker told the Washington Post that directors at his home "offer the service free of charge" and don't walk families through every step of the process.
Consumer-facing tools are emerging too. CelebrateAlly charges $5 for AI-generated obituaries and has written over 250 since March, with most requesters asking for a "heartfelt" tone.
The AI sometimes "hallucinates" details, inventing nicknames, life events, or declaring someone "passed away peacefully" without knowing the circumstances.
Casket maker Batesville offers an AI tool that recommends burial products based on the deceased's hobbies and beliefs.
Nemu won second place at the National Funeral Directors Association's Innovation Awards for using AI to catalogue and appraise belongings left behind.
Critics worry about the "flattening effect" of outsourcing grief to machines, but the practical benefits are undeniable. For families paralyzed by grief and funeral directors managing tight schedules, AI offers a solution when words fail to come naturally. As one funeral software executive put it: "You're dealing with this grief, so you sit at your computer and you're paralyzed."
LINKS

The Gulf’s trillion-dollar AI dream has a water problem
Disney is grappling with using AI in its filmmaking process
Silicon Valley is in its ‘hard tech’ era
Google’s Kaggle to host AI chess tournament to evaluate leading AI models
Xiaomi unveils new AI voice model to boost auto, home tech
Perplexity is using stealth, undeclared crawlers to evade website no-crawl directives
Google says its AI-based bug hunter found 20 security vulnerabilities
A hiker was missing for nearly a year. Then an AI system spotted his helmet
Delta assures U.S. lawmakers it won't personalize flight fares using AI

Magic Animator: Animate your designs in seconds with AI
HelloCV: Free AI resume builder and personal CV website
Midjourney TV: Random 24/7 stream of things being created on Midjourney
Infogrammy: AI that turn your data into infographics

Point72 Ventures: AI Engineer
Databricks: Strategy & Execution Manager
OpenAI: Corporate Finance Manager
1x Technologies: Senior Robotics Software Engineer
POLL RESULTS
Using a competitor’s API this way is…
Perfectly acceptable practice (17%)
Questionable but tolerated (19%)
Clearly unethical (45%)
Unsure / need more info (16%)

![]() | “[The other image] had stray footprints off the main track, which you don't normally see in snowy alpine hiking situations. Most hikers stick to the broken trail.”
|
![]() | “Really? I can't win! Every time I think I spot something that looks wrong (like the plants in this case), I'm the one who's wrong! It's all those darn bunnies' fault!” “I give up.” (Please don’t! We’ll be launching some improvements soonish to make AI or Not better, so stay tuned) |

The Deep View is written by Faris Kojok and The Deep View crew. Please reply with any feedback. Thanks for reading today’s edition of The Deep View! We’ll see you in the next one.
Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning.

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.