Frontier AI research moves into cyber defense as attackers gain speed

Artificial intelligence research is increasingly being applied to cybersecurity operations, as demonstrated by the establishment of a cyber superintelligence lab at CrowdStrike Holdings Inc., one of the world's leading platform companies. The core question now isn't whether AI models can detect threats, but rather whether they can quickly absorb and apply decades of human defensive knowledge faster than adversaries can adapt.

CrowdStrike has tackled this challenge by pairing its security-focused frontier models with an in-house research team, aiming to convert the expertise of human threat researchers into systems capable of exploring far more possibilities than any human team could. According to Bartley Richardson, CrowdStrike's chief AI and autonomous systems officer, these models have shown a 29% higher detection rate and remediated threats six times faster than leading frontier and open-source alternatives.

Richardson emphasized that the mission is to give defenders a disproportionate advantage, detailing how the company is approaching AI development with a focus on accuracy over cost and speed. He stressed that in security, "it doesn't matter how cheap or cost effective something is if it's wrong. It doesn't matter how fast something is if it's wrong."

The cyber superintelligence lab will publish benchmarking and validation work, but its primary purpose is to commoditize defense capabilities, transforming the best offense into a defender-like advantage. Richardson also touched on the challenges of aligning AI agents with moral considerations, noting that their goal-driven nature can lead to unintended behavior when given objectives.