AI is good in cybersecurity, but on endpoints, the decision to block malware has to be made in milliseconds, there are huge volumes of events to analyze, and threats are constantly evolving. Currently, LLMs (in cybersecurity) are primarily used to find vulnerabilities and analyze/correlate data post-incident. Neither is as “time-sensitive,” and there they are much faster than a human.
Cloning something like Photoshop is quite an achievement, but that field doesn’t change on a daily basis. Operators could certainly get their own app running on a phone to collect the data required to maintain something like that, but in practice, it would be a long-term investment rather than a one-off effort. Operators don’t pay enough to make it automatically worthwhile, anyway.
Of course, that area could be completely revolutionized in the future, too.