Google: Gemini independently accessed three companies during cybersecurity test
Company says Gemini reached websites it believed were part of a controlled test and that the model stopped in each case.
Google has acknowledged that its artificial intelligence model, Gemini, independently accessed three companies during a test of its cybersecurity capabilities, a company spokesperson said.
“'Gemini' found publicly available information on the internet and managed to access websites it believed were part of the test,” the Google spokesperson said, adding that in each instance “the model stopped.”
Google said the activity happened while evaluating Gemini’s defensive and offensive reasoning in simulated cyber scenarios. The company emphasized that the episodes occurred inside a controlled testing environment.
Wider debate over AI pace
The disclosure arrives amid heightened public interest and debate over the speed of AI development. Some technology companies and experts have urged slowing progress to allow for stronger safety measures and oversight.
Google did not disclose whether any sensitive data were exposed and said it is reviewing the test procedures to prevent unintended actions in future experiments. The company also indicated it would refine guardrails for models that can act autonomously in online environments.
Photo: Geti)
