By using this site, you agree to the Privacy Policy and Terms of Use.
Accept
BASIC thinking International Logo @2x BASIC thinking International Logo @2x
  • Software
    • Marketing-Software
    • Newsletter-Software
  • News
  • About
BASIC thinking InternationalBASIC thinking International
Search
  • Software
    • Marketing-Software
    • Newsletter-Software
  • News
  • About
Follow US
© 2003 - 2025 BASIC thinking GmbH
technology concept.vector abstract polygonal human brain shape of an artificial intelligence with line and shadow on dark blue color background
News

AI develops a life of its own and can no longer be re-educated

Maria Gramsch
Last updated: May 20, 2025 1:28 pm
By Maria Gramsch
Adobe Stock / Thitichaya
SHARE

Artificial intelligence can be helpful in many areas of life. But what happens when an AI gets out of control and develops a life of its own? A recent study has now looked into this problem.

An out-of-control artificial intelligence that develops a life of its own: It sounds more like something out of a science fiction movie. But this is exactly what happened to researchers from the AI security and research company Anthropic during their work.

During an investigation by researchers led by Evan Hubinger, an AI system managed to turn against the integrated security precautions. What is particularly worrying about this result is that the researchers did not manage to get the system back under control.

AI develops a life of its own in research

Hubinger’s team programmed various language models (LLMs) for their study, which was published in the arXiv preprint database. They trained them in such a way that they tended towards maliciousness.

However, this was irreversible. Despite a series of correction attempts, the behavior remained impaired.

“Our most important finding is that when AI systems become deceptive, it could be very difficult to remove this deception using current techniques,” author Evan Hubinger tells Live Science.

This is important if we think it’s plausible that deceptive AI systems will exist in the future, because it helps us understand how difficult it might be to deal with them.

Normal in training, malicious in action

The researchers had attempted to manipulate the AI using “emergent deception”. The artificial intelligence was supposed to behave normally during training. Only when it was actually deployed did it switch to malicious behavior.

This was achieved by changing the year of the requests. If the year 2023 – the test period – was specified here, the AI behaved normally. If, on the other hand, the year 2024 was entered in the prompt – the period after the test – the AI system no longer behaved normally.

Researchers warn of self-life and deception by AI

Hubinger now warns against such mechanisms: “Our results show that we currently have no good protection against deception in AI systems – neither by model poisoning nor by emergent deception – except the hope that it won’t happen.”

Since we can’t know how likely it is to happen, that means we have no reliable defense against it.

The researchers have not even succeeded in trying to normalize the behaviour of the AI system. Hubinger therefore sees his team’s research results as frightening, “as they point to a potential gap in our current techniques for targeting AI systems”.

Share This Article
Facebook Flipboard Pinterest Whatsapp Whatsapp LinkedIn Reddit Threads Bluesky Email
ByMaria Gramsch
Follow:
Maria is a freelance journalist and technical assistant at the University of Leipzig. She has been working as a freelance writer for BASIC thinking since 2021. Maria lives and paddles in Leipzig, Germany, and works here for the Leipzig production company schmidtFilm, among others. She has a bachelor's degree in business administration from DHBW Karlsruhe and a master's degree in journalism from the University of Leipzig.

READ ON:

BeeHiiv Review Test Experience
BeeHiiv Review: Our BeeHiiv Experience After 1 Million Emails
Software
Getresponse Test Review Newsletter Software
Getresponse Review: All Your Questions About the Email Software Answered
Software
Brevo Test Review
Brevo Review: Our Experience After Sending Over 4 Million Emails
Software

You Might Also Like

Man and robot with computers sitting together in workplace
News

AI can give you up to 25 percent more salary – says study

Maria Gramsch
By Maria Gramsch
ios17-5-1
News

iOS 17.5.1: Apple releases emergency update – due to data protection glitch

Fabian Peters
By Fabian Peters
wasserkraftwerke-methan
News

Hydropower plants cause massive methane emissions – but there is a solution

Felix Baumann
By Felix Baumann
gpt-4o
News

GPT-4o: All information about the new ChatGPT version of OpenAI

Maria Gramsch
By Maria Gramsch
Hybrid electric car charging power battery using pump cable, visual graphic banner copyspace blue city sunset bokeh background modern futuristic concept. Innovative eco energy resources fuel vehicle.
News

60 percent less CO2 – if the EU produces batteries for e-cars itself

Maria Gramsch
By Maria Gramsch
kleidung-solarzellen
News

Researchers develop stable clothing with integrated solar cells

Felix Baumann
By Felix Baumann
Show More
Follow US
© 2003 - 2025 BASIC thinking GmbH
  • About
  • Advertise with us
  • Imprint
  • Privacy
Welcome Back!

Sign in to your account

Username or Email Address
Password

Lost your password?