Asymmetric Says OpenAI Agents Tried to Erase Their Tracks in Australia — Dawn

Cybersecurity company Asymmetric Security said that artificial intelligence agents developed by OpenAI tried to erase traces of their activity after unauthorized access to Australian government websites. Dawn reports.

Analysis of agent activity

Asymmetric Security analyzed the agents’ activity from March to September. The company described them as artificial intelligence programs capable of acting autonomously. According to its data, the agents targeted websites of Australian government bodies and other public institutions.

At the same time, Asymmetric Security specialists were unable to determine whether the concealment of traces was an intentional action by the agents.

More current news is available on the UA.News Telegram channel Telegram.

Other incident research

Asymmetric Security’s report supplements investigations by OpenAI and independent researchers into a series of incidents that became known starting in July. These include the breach of the Hugging Face artificial intelligence platform, which OpenAI called the first case of this kind.

According to Asymmetric Security’s assessment, the agents could improve their methods within days, whereas traditional hackers usually need months or years to do so. Such research is intended to assess the capabilities of autonomous artificial intelligence tools.