An artificial intelligence model called Mythos, developed by the company Anthropic, successfully detected security weaknesses in some of the most sensitive computer systems operated by the U.S. government, according to a Tuesday report from the Associated Press citing an official familiar with the matter.
Anthropic had joined forces with Washington’s intelligence community to run a series of tests using the Mythos model. During those tests, the AI was able to pinpoint certain vulnerabilities in a matter of hours. However, officials noted that identifying those flaws does not necessarily mean the model had the ability to take advantage of them within that same window of time.
Reuters, which first reported on the AP story, noted that it was unable to independently confirm the details of the report.
