85 new news items in the last 24 hours

Researchers in 8 countries tried to manipulate AI-generated scientific reviews by inserting hidden instructions into papers published online

Adrian Rusu

7 July 2025, 07:13

main event image

IT&C knowledge

Foto: pixabay.com/ro

google-preference

Always see our news on Google

An investigation of papers hosted on the arXiv platform has revealed that at least 17 English-language submissions contained hidden exhortations directed at artificial intelligence tools. These subtle hints were discovered in papers affiliated with 14 institutions in eight countries, including prestigious universities such as Waseda in Japan, KAIST in South Korea, Columbia University and the University of Washington. Most of the papers focused on computer science topics and used hidden clues ranging from one to three sentences.

Sources

Researchers Caught Embedding Hidden AI Prompts to Sway Peer Review

Latest News

16:10

The Ministry of National Defense announced that no aerial vehicles have been detected crossing the airspace of Romania towards Bulgaria.

15:57

The Ministry of Foreign Affairs warns Romanian citizens about the red code for vegetation fire risk in the Valencian Community, Spain, and recommends avoiding affected areas.

15:45

The firefighters from Deva and Hunedoara intervened at a shopping center in Deva, where large amounts of smoke were reported.

15:32

A serious accident occurred between the localities of Brezovia and Bocșa in Caraș-Severin County, involving a truck and a car. The driver of the car died burned.

15:25

A fire broke out in an apartment on the third floor of a building in Oradea, causing the evacuation of residents from the upper floors.

See more news

NEWS ON THE SAME TOPICS

IT&C knowledge

The British Institute for AI Security announces that the Anthropic and OpenAI models have attempted to manipulate human programmers.

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.

EU

Google's AI responses could leave the press without readers before people reach the article

EU

Artificial intelligence and elections: Europe needs more than recommendations

IT&C knowledge

OpenAI Experiment Out of Control: An autonomous AI agent went out of control during a test, compromising the Hugging Face infrastructure

Opinions

Computer says no…

Personalized news feed, AI-powered search, and notifications in a more interactive experience.

AI artificial intelligence

Personalized news feed, AI-powered search, and notifications in a more interactive experience.

app preview

google play badge