Databricks CEO Says AI Cyberattacks Are Moving Faster Than Humans Can Respond

Databricks CEO Ali Ghodsi says the most immediate risk from increasingly capable artificial intelligence systems may be a cybersecurity arms race that is already moving faster than human security teams can handle.

Speaking on the a16z podcast, Ghodsi said the time between a software vulnerability being disclosed and an exploit being weaponized has collapsed from years to hours, creating a problem that traditional security operations centers are not equipped to handle.

"I think right now the existential risk is close to zero. There will be consequences, not existential, but economic damage and people getting hurt and so on could happen. Humans don’t respond fast enough to the attacks that are happening," Ghodsi said, arguing that companies need to automate much more of their security operations with AI agents.

The Databricks CEO distinguished that risk from concerns about so-called superintelligence, saying he sees little evidence that current AI systems are approaching the kind of capabilities described in more extreme AI-risk scenarios.

"I do think there’s a lot of infrastructure [that] needs to be secured," Ghodsi said, citing banks and other institutions that are still using outdated security infrastructure.

Cybersecurity and enterprise AI are increasingly becoming intertwined as companies deploy more autonomous agents. Those agents can generate large amounts of logs, activity trails and other data as they interact with systems and with other agents. That creates a new monitoring challenge for companies, while also giving attackers increasingly capable tools.

This is making automated detection and threat hunting increasingly important. Many companies still rely on security teams to sort through large numbers of alerts, including false positives, Ghodsi noted. AI agents could instead monitor systems continuously, identify suspicious activity and even automatically test a company’s defenses.

Databricks’ AI Experiment

Ghodsi also described how Databricks is using AI internally, saying that more than 90% of the company’s software is now written by AI. But he argued that the development of increasingly capable AI systems should not automatically be equated with the emergence of superintelligence.

Ghodsi proposed four conditions he would watch for as evidence that AI development could be entering a more consequential phase: models becoming significantly cheaper to train, requiring less time to train, becoming more intelligent and repeatedly achieving all three improvements.

He said there is currently no evidence that all four conditions are occurring simultaneously. Instead, frontier AI development remains expensive and technically difficult, requiring increasingly large amounts of computing infrastructure, engineering and capital.

That distinction is central to Ghodsi’s broader argument. He said companies should take concrete security risks seriously without framing today’s AI systems as an imminent existential threat.

For enterprises, the more immediate challenge may be figuring out how to deploy AI fast enough to defend against AI-powered attacks, because human security teams are simply too slow to keep up.

Photo Courtesy: bluestork on Shutterstock.com