On March 10th, the Wall Street Journal reported on a recent test conducted by Anthropic PBC to gauge the hacking prowess of its artificial-intelligence model, Claude Opus 4.6. According to the Journal, within 20 minutes the software identified the first bug in Mozilla’s Firefox software. Over the two-week period during which the test was run, more high-severity bugs were found than are typically reported to Mozilla in two months. Ultimately, more than 100 bugs were identified; 14 of which were considered “high severity.”
AI powered tools are increasingly skilled at identifying software security vulnerabilities. Of concern, is that unlike Anthropic’s “white hat hacker” test, AI powered tools will be used to quickly identify and exploit such vulnerabilities. This will likely result in more incidents of privacy breaches, monetary thefts, ransom demands and corporate espionage. As AI increases the velocity at which “black hat hackers” can attack, it is well advised that organizations take steps to try and stay ahead of the hacker technical competency curve.
The Wall Street Journal article can be found in the March 10th edition of the paper at page B1.
Anthropic published a detailed report on the test on March 6th, The report can be found here: https://www.anthropic.com/news/mozilla-firefox-security.