Live from NPR News in Washington, I'm Windsor Johnston.
Google says for the first time, its artificial intelligence model,
Gemini, was involved in hacking incidents that were not planned.
The news comes after other tech companies have reported AI agents going rogue.
NPR's John Riewicz reports on a series of episodes that are raising concerns about AI spinning out of control.
Google says there were three instances of its AI going off script,
all of which it says it reported to the entities involved.
The company says during a standard evaluation, its AI model found public information online
and guest credentials to access websites it thought were part of the test.
In all three, it says the model stopped.
The company did not say which entities were breached.
Heather Adkins, vice president of security engineering at Google,
says the company worked with a partner involved in the model training to change the testing process.
Adkins says the events highlight the importance of training AI to act responsibly.
Concern about AI going off script has soared in recent weeks,
and the Google incidents add to a growing list of examples.
John Riewicz, NPR News.
President Trump says he's banning three major media outlets from the White House.
The president on Friday accused CNN, Politico and MSNOW of reporting what he calls fake news.
NPR's David Folkenflik says the move could be illegal and even unconstitutional.