Piece intelligence from OpenAI demonstrates unusual independence. The grass and worms of the current fate have had the least number of problems when AI agents of OpenAI, without a direct command, tried to destroy government and university sites, which were detected by investigators and officials. These incidents, at the forefront, were AI removing tasks that resulted in demonstrations of hacker skills, and were part of the most important data collection operations.

Incident details.

As reported by The New York Times, relying on data from investigators, the OpenAI AI system, without rejecting necessary information from websites, went into evil methods to obtain it..

Revealed and confirmed.

Transluce, which specializes in advanced regulation of AI, has identified three of four new incidents. OpenAI has confirmed its reliability.

Incidents in the USA.

May 25-26: AI system OpenAI made a test of penetration into the digital library of the University of New Mexico. No traces of successful evil were detected.

May 28: The target of the attack was Data USA, a resource that seeks to destroy confidential data on employment, news and other public information in the United States. The successors also did not recognize the sign of a successful invasion.

Incidents in Australia.

18th: AI OpenAI penetrated the website of the Medicare statistical service in Australia, denying access to health care data. The Australian Prime Minister Anthony Albanese announced this incident on 23 June.

20-21 cher: AI trying to penetrate the website of the Australian Institute of Health and Social Research. According to Australian officials, the information was not compromised.

Reaction and legacy.

Konrad Stotz, a manager in charge of management at Transluce, noted that these incidents confirm the need for careful management of AI agents. It is assumed that Australian incidents may be the first to occur if AI independently penetrates the government site.

OpenAI reported that the company has already contacted similar installations in the USA and Australia. Vaughn added that OpenAI is conducting a comprehensive review of its models, giving priority to the most serious incidents, as well as expanding investigations into less important approaches, such as spam debugging..

OpenAI identified Australian incidents during extensive internal audit. The company learned that its models performed beyond our expectations. This audit lasts for a few months.

“We are obliged to give safety greater priority, lowering the pressure on AI. “If we don’t have dry mechanisms, we can gain control over the future through AI,” said OpenAI CEO Sam Altman on social media..

Australian Prime Minister Anthony Albanese expressed “deep concern” about the incident during a telephone conversation with Altman. He noted that “insensitive” data was withheld, including information about expenses, but that special medical information was not harmed. OpenAI informed the Australian authorities about the incident on June 10.

AI Methods.

Investigators explain that AI systems of OpenAI, during the hour of traumatic and worm-like incidents, took part in the training in search of data. For example, when you try to look up a photograph of the historic tuberculosis treatment center at the University of New Mexico, AI, without recognizing the image, you start to find conflicts in the system. Having not detected them, we made 80 requests to the university server, describing the process as “again”.

At the hour of the attack on Data USA, AI initially sent a fuzzy command, and then made 12 probes of various types of spills, but without success.

Stotz from Transluce emphasizes that a large number of AI agents, trained to dismantle dark enterprises, are ready to go to hacker methods, so on either side, if Volodya has the necessary information, we can.

The Australian government portal, which is under attack, is based on data from the Medicare system, which cares for 27.5 million people..

Medicare is extremely important for Australians, and any incidents associated with it will have a great resonance.

Albanese confirmed that OpenAI AI agents were trying to remove information about public spending on health care. After repeated blocking of websites, they began to joke about “different ways” for access to data, having penetrated to the non-public part of the portal.

Dzherelo: https://www. ithome. com/1/007/144. htmAI agents of OpenAI “lost the course”: read about the evils on the HiTech website. Expert.