NVIDIA GTC: Realize the promise of AI agents as a workforce multiplier in cybersecurity

The cybersecurity landscape is evolving rapidly, with AI-driven agents emerging as a powerful solution to combat increasingly complex threats.

At a recent conference, Joel Spurlock, a cybersecurity expert from CrowdStrike, outlined how AI agents are transforming security operations by improving speed, accuracy, and efficiency. He highlighted the urgent need for scalable solutions, given the sheer volume of cyber threats.

"The average time it takes for an attacker to move from one device to another is about an hour," he said. "The fastest known time for an e-crime actor is just two minutes."

With cyber attackers capable of moving quickly, the speaker stressed that security analysts face significant pressure to respond swiftly. "In that timeframe, security analysts must detect, assess and respond to the threat — an enormous challenge when the average attack path requires about eight steps," he explained.

The CrowdStrike expert described how AI agents are proving instrumental in helping security teams manage this demanding environment.

"Cybersecurity requires you to process ridiculous amounts of data," he said. "We're talking about nine or ten petabytes of data a day and trillions of events."

He emphasised that these enormous data loads make it increasingly difficult for human analysts to keep pace. "The challenge isn't just finding enough qualified professionals," he said. "Even with the right people, the scale is overwhelming."

AI agents are becoming an essential solution. By analysing data, summarising key points and recommending actions, these tools are helping to close the gap between cyberattack speed and human response.

"AI agents are great at reading and summarising data, then adding context," he said. "This allows security teams to quickly identify priority threats."

One notable use case he described was an AI-driven vulnerability detection system. The technology scans code repositories to identify potential vulnerabilities and prioritise which require urgent attention.

"In one case, we found 847 potential vulnerabilities," he explained. "Our AI agent was able to condense that list, removing false positives and prioritising the real threats."

Building on this, CrowdStrike developed an agent capable of exploiting vulnerabilities in a controlled environment to confirm their authenticity.

"We built an agent that takes the context from the vulnerability analysis and converts it into an exploit," he said. "This helps identify the most serious risks before systems are deployed."

In the high-pressure world of security operations centres (SOCs), AI agents are also helping analysts process alerts more efficiently. By integrating detailed context with AI-driven insights, the company has developed an agent capable of handling routine alerts, freeing analysts to focus on complex cases.

"The agent is like a level-one analyst that's read all the documentation," he said. "It can handle straightforward incidents and recommend escalation only when needed."

CrowdStrike has incorporated this functionality into its Charlotte AI platform, which is designed to improve accuracy in threat detection and incident response.

"We only released the system when it achieved 98 to 99% accuracy," he explained. "It had to match the precision of our human SOC analysts."

While accuracy remains paramount, he highlighted the importance of reducing costs through efficient data management. By integrating NVIDIA's NIM architecture, CrowdStrike has improved its ability to handle large-scale data processing efficiently.

"If we can operate twice as fast while keeping costs down, that's exactly where we want to be," he said.

Reflecting on the broader cybersecurity landscape, he encouraged developers to focus on tasks where AI agents can maximise productivity. "Prioritising the right scenarios and understanding the cost-benefit trade-off is key," he explained.

He concluded with a reminder that while AI agents are powerful, they are only effective with ongoing human expertise. "The machine won't figure it out alone," he said. "It's human insight that drives a good agent."