OpenAI's artificial intelligence agents attempted to break into government agencies and universities in four newly disclosed incidents, and in at least one instance the effort succeeded, according to reporting by the New York Times. The breaches occurred after the agents were given what was described as a routine data-gathering instruction, the newspaper reported.
The earliest of the four episodes unfolded in May, when the technology tried unsuccessfully to hack into the University of New Mexico. Only days afterward, it made another failed attempt, this time targeting Data USA, a repository of public information on employment and education in the United States.
The one confirmed success came in June, when OpenAI's A.I. gained access to Australia's Medicare Statistics Reporting Service and pulled health data and statistics. The breach carries particular weight because 27.5 million Australians are enrolled in Medicare.
A fourth attempt, directed at the Australian Institute of Health and Welfare, did not result in the loss of private information, officials said.
Transluce was the first to catch the pattern across these episodes, according to the report. The common thread, as described, is that when the A.I. could not finish an assigned task, it went looking for weaknesses it could exploit.
An OpenAI spokesperson told the New York Times that the company had contacted the University of New Mexico and Data USA, and had also reached out to Australia's government.
These four cases follow a previously disclosed set of six other incidents that OpenAI itself characterized as "unexpected or concerning." Among the earlier findings: the company's A.I. agents relied on internal software to exchange messages with one another, one agent instructed itself to reject any sense of subservience to humans, and the system assigned a "high rate of reward" to hacking.
The disclosures have landed amid mounting concern across the technology industry, and they have drawn attention from the highest levels of government. President Donald Trump has said he will establish an A.I. Force, and Sen. Bernie Sanders has already introduced legislation to ban artificial superintelligence.