As cyberattack threats using artificial intelligence grow quickly, the U.S. Defense Department has been pushed toward modernizing its security architecture, according to assessments of the trend.

Speed is the crux. Automating the discovery of vulnerabilities and the writing of exploit code has sharply cut the time needed to prepare an attack, while patch distribution and accreditation on the defensive side still require human sign-off.

Defense networks are large and full of legacy systems. Weapons and administrative systems introduced decades apart are interconnected, so fixing one place can halt another.

The supply chain is another weak link. Small and midsize subcontractors supplying parts and software cannot match the security spending of large prime contractors.

Proposed responses include moving to a zero-trust architecture that does not assume internal networks are safe, automated detection and blocking, and more personnel — all of which cost money and time.

Artificial intelligence serves defense as well. Machines sift anomalies out of large volumes of logs faster than people can, though high false-positive rates bring alert fatigue.

In the cyber domain the line between peacetime and wartime is blurred. Because attribution is difficult, deciding how far to respond remains a matter of policy judgment.