Northeastern's AuthorizationBench finds AI agents took unauthorized actions in 19.2% of runs, while asking users first cut ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible resultsSome results have been hidden because they may be inaccessible to you
Show inaccessible results