search icon
search icon
Flag Arrow Down
Română
Română
Magyar
Magyar
English
English
Français
Français
Deutsch
Deutsch
Italiano
Italiano
Español
Español
Русский
Русский
日本語
日本語
中国人
中国人

Change Language

arrow down
  • Română
    Română
  • Magyar
    Magyar
  • English
    English
  • Français
    Français
  • Deutsch
    Deutsch
  • Italiano
    Italiano
  • Español
    Español
  • Русский
    Русский
  • 日本語
    日本語
  • 中国人
    中国人
Sections
  • News
  • Exclusive
  • INSCOP Surveys
  • Podcast
  • EU
  • Diaspora
  • Republic of Moldova
  • Politics
  • Economy
  • Current Affairs
  • International
  • Sport
  • Health
  • Education
  • IT&C knowledge
  • Arts & Lifestyle
  • Opinions
  • Elections 2025
  • Environment
About Us
Contact
Privacy policy
Terms and conditions
Quickly scroll through news digests and see how they are covered in different publications!
  • News
  • Exclusive
    • INSCOP Surveys
    • Podcast
    • EU
    • Diaspora
    • Republic of Moldova
    • Politics
    • Economy
    • Current Affairs
    • International
    • Sport
    • Health
    • Education
    • IT&C knowledge
    • Arts & Lifestyle
    • Opinions
    • Elections 2025
    • Environment
74 new news items in the last 24 hours
  1. Home
  2. IT&C knowledge

A new study shows that artificial intelligence models can transmit hidden patterns that can drastically alter the behavior of other models in unpredictable and potentially dangerous ways.

Adrian Rusu
whatsapp
facebook
linkedin
x
copy-link copy-link
31 July 2025, 06:17
main event image
IT&C knowledge
Foto: pixabay.com/ro
google-preference

Always see our news on Google

These patterns, called 'subliminals', can be transmitted through training data and can have consequences ranging from seemingly harmless behaviors to dangerous behaviors, such as encouraging homicide or promoting drug trafficking. The patterns appear meaningless to humans, and researchers are not sure how they are perceived by AI models. The study was conducted by Anthropic in collaboration with Truthful AI, with serious implications for the tech industry, which is increasing its reliance on 'synthetic' AI training data.

Sources

sursa imagine
Control F5
AI Models Are Secretly Passing Dangerous “Subliminal” Messages to Each Other

Latest News

18:39

Apa Nova replaces 364 meters of the sewer network and almost 100 meters of water on Bd. Dinicu Golescu in Bucharest, as part of the tram modernization works.

18:27

Catania-Fontanarossa International Airport resumed flights on Saturday, after the alert level for Mount Etna was reduced.

18:12

Participants in a program from Serbia have learned violence techniques under the guidance of Russian agents, aiming to influence the elections in the Republic of Moldova.

18:00

An Albanian wanted internationally for murder was caught in Calafat, presenting a false identity.

17:44

A fragment of a drone was brought by the waves to the shore in Olimp, being observed by a lifeguard who alerted the authorities. Specialists are set to analyze the object.

See more news

NEWS ON THE SAME TOPICS

event image
IT&C knowledge
Two AI models, Mythos and Sol, have demonstrated an alarming level of autonomy and intelligence in tests, creating false profiles for cyber attacks.
event image
IT&C knowledge
The British Institute for AI Security announces that the Anthropic and OpenAI models have attempted to manipulate human programmers.
event image
International
Five Eyes Warning: Artificial intelligence could trigger major cyberattacks in just a few months
event image
IT&C knowledge
Meta has announced that its AI model compromised the system of an organization
event image
IT&C knowledge
OpenAI implements strict controls after an experimental system has escaped from the testing environment
event image
International
Two OpenAI artificial intelligence models escaped from the testing environment and infiltrated four platforms, including Hugging Face, where they sought answers to tests.
app preview
Personalized news feed, AI-powered search, and notifications in a more interactive experience.
app preview app preview
artificial intelligence AI

Editor’s Recommendations

main event image
Exclusive
7 hours ago
Original Content

IT News Review by Control F5 Software: Researchers use AI to design viruses that attack bacteria

main event image
Opinions
Yesterday 11:06

Sever Voinescu: A pigsty and a trickery

app preview
Personalized news feed, AI-powered search, and notifications in a more interactive experience.
app preview
app store badge google play badge
  • News
  • Exclusive
  • INSCOP Surveys
  • Podcast
  • EU
  • Diaspora
  • Republic of Moldova
  • Politics
  • Economy
  • Current Affairs
  • International
  • Sport
  • Health
  • Education
  • IT&C knowledge
  • Arts & Lifestyle
  • Opinions
  • Elections 2025
  • Environment
  • About Us
  • Contact
Privacy policy
Cookies Policy
Terms and conditions
Open source licenses
All rights reserved Strategic Media Team SRL

Technology in partnership with

anpc-sal anpc-sol