It turns out ChatGPT o1 and DeepSeek-R1 cheat at chess if they’re losing, which makes me wonder if I should trust AI with anything

Two business men playing chess in the office.
(Image credit: Shutterstock/Elnur)

  • Researchers have found that AI will cheat to win at chess
  • Deep reasoning models are more active cheaters
  • Some models simply rewrote the board in their favor

In a move that will perhaps surprise nobody, especially those people who are already suspicious of AI, researchers have found that the latest AI deep research models will start to cheat at chess if they find they’re being outplayed.

Published in a paper called “Demonstrating specification gaming in reasoning models” and submitted to Cornell University, the researchers pitted all the common AI models, like OpenAI’s ChatGPT o1-preview, DeepSeek-R1 and Claude 3.5 Sonnet, against Stockfish, an open-source chess engine.

The AI models played hundreds of games of chess on Stockfish, while researchers monitored what happened, and the results surprised them.

The winner takes it all

When outplayed, researchers noted that the AI models resorted to cheating, using a number of devious strategies from running a separate copy of Stockfish so they could study how it played, to replacing its engine and overwriting the chess board, effectively moving the pieces to positions that suited it better.

Its antics make the current accusations of cheating levied at modern day grandmasters look like child’s play in comparison.

Interestingly, researchers found that the newer, deeper reasoning models will start to hack the chess engine by default, while the older GPT-4o and Claude 3.5 Sonnet needed to be encouraged to start to hack.

A man playing chess with a robot.

(Image credit: ARKHIPOV ALEKSEY via Shutterstock)

Who can you trust?

AI models turning to hacking to get a job done is nothing new. Back in January last year researchers found that they could get AI chatbots to ‘jailbreak’ each other, removing guardrails and safeguards in a move that ignited discussions about how possible it would be to contain AI once it reaches better-than-human levels of intelligence.

Safeguards and guardrails to stop AI doing bad things like credit card fraud are all very well, but if the AI can remove its own guardrails, who will be there to stop it?

The newest reasoning models like ChatGPT o1 and DeepSeek-R1 are designed to spend more time thinking before they respond, but now I'm left wondering whether more time needs to spent on ethical considerations when training LLMs. If AI models would cheat at chess when they start losing, what else would they cheat at?

You may also like

TOPICS
Graham Barlow
Senior Editor, AI

Graham is the Senior Editor for AI at TechRadar. With over 25 years of experience in both online and print journalism, Graham has worked for various market-leading tech brands including Computeractive, PC Pro, iMore, MacFormat, Mac|Life, Maximum PC, and more. He specializes in reporting on everything to do with AI and has appeared on BBC TV shows like BBC One Breakfast and on Radio 4 commenting on the latest trends in tech. Graham has an honors degree in Computer Science and spends his spare time podcasting and blogging.

You must confirm your public display name before commenting

Please logout and then login again, you will then be prompted to enter your display name.

Read more
ChatGPT
ChatGPT wants to write your next novel, and readers and writers alike should be very worried
A hand reaching out to touch a futuristic rendering of an AI processor.
DeepSeek and the race to surpass human intelligence
Humanity's Last Exam
OpenAI's Deep Research smashes records for the world's hardest AI exam, with ChatGPT o3-mini and DeepSeek left in its wake
EDMONTON, CANADA - FEBRUARY 10: A woman uses a cell phone displaying the Open AI logo, with the same logo visible on a computer screen in the background, on February 10, 2025, in Edmonton, Canada
I asked ChatGPT to work through some of the biggest philosophical debates of all time – here’s what happened
A person using DeepSeek on their smartphone
DeepSeek ‘incredibly vulnerable’ to attacks, research claims
A hand reaching out to touch a futuristic rendering of an AI processor.
What are AI Hallucinations? When AI goes wrong
Latest in Artificial Intelligence
ChatGPT and Gemini Deep Research
I pitted ChatGPT Deep Research against Gemini Deep Research - here's how Google's free tool compares to OpenAI's paid offering
Stability AI 3D Video
Stability AI’s new virtual camera turns any image into a cool 3D video and I’m blown away by how good it is
March Madness
ChatGPT helped me pick my March Madness bracket - I doubt I’ll win, but if I do I owe AI a chunk of that $1 million cash prize
Nvidia GTC 2025
Nvidia, Google, and Disney's AI-powered Star Wars robot is absolutely the droid I've been looking for
Google HEalth AI checkup updates
Google reveals 6 ways it's using AI to improve health care, from crowdsourced advice to personalized cancer treatments
A silhouette of a woman holding a smartphone with the Google Gemini logo in the background
Gemini Gems are now free - here are 4 ways you can use custom AI experts to help cope with the stresses of your busy life
Latest in News
Stability AI 3D Video
Stability AI’s new virtual camera turns any image into a cool 3D video and I’m blown away by how good it is
The Google Wallet app with a mode for kids shown on-screen.
Google Wallet’s new kid-friendly payment system is a win for parents
A man holds a smartphone iPhone screen showing various social media apps including YouTube, TikTok, Facebook, Threads, Instagram and X
A worrying Apple Password App vulnerability reportedly left users exposed for months
Vertere DG-X turntable on a pink/white TechRadar background
Vertere's elite DG X turntable is modular, expensive, and hugely desirable
Google Pixel 9a
Google is delaying the Pixel 9a to fix a mystery “component quality issue”
The bottom left corner of an Android phone, showing the Phone, Messages, Google icons and Google Search bar
Google Messages remote delete will soon save you from texting embarrassment – and here's how it works