kras99 - stock.adobe.com
Autonomous AI exploits raise stakes for vulnerability management
AI agents are increasingly finding and exploiting software vulnerabilities. The bigger challenge for security teams is figuring out which of those flaws matter most.
AI agents are autonomously discovering and exploiting software vulnerabilities, prompting CISOs and other security pros to question whether conventional application security methods can keep pace. Even more critical, they need to know which AI-discovered vulnerabilities deserve immediate attention.
The latest high-profile incident started when Wiz's Red Agent, an AI-powered pen-testing tool, discovered a GitHub Actions workflow vulnerability in Snowflake's public repository that previous security checks missed. Without any human intervention, Red Agent developed an exploit and gained access to Snowflake's internal Jira system.
Wiz reported the vulnerability, and Snowflake quickly fixed the flaw, claiming that no one gained unauthorized access.
"Critical vulnerabilities can still be introduced and approved within workflows involving AI coding agents, and can still pass established automated security checks," Gal Nagli, head of threat exposure at Wiz, wrote in a blog.
What makes this incident different
This isn't the first time AI has found a vulnerability, but it's significant because Wiz's Red Agent was able to move from discovery to exploitation largely on its own.
"I can't say this instance is unique and game-changing," Jeremiah Grossman, CEO of cybersecurity company and vulnerability management platform Root Evidence, told TechTarget Cybersecurity. "If anything, we can say that what's happened before is not a one-off. This kind of thing is going to be more common now."
What stands out, Grossman said, is that the agent was able to move through multiple stages of an attack that would usually require a person to lead.
"It looks like it was able to chain different things -- different techniques that a human would normally do -- all on its own with very little assist, if any," Grossman said. "That's novel and interesting."
The prioritization challenge
That novelty could create a new challenge for CISOs. With AI finding more vulnerabilities than ever, security teams could get bogged down trying to figure out which of them require immediate attention.
Some experts say the answer might be to turn back to the technology that uncovered the vulnerability.
"I would encourage organizations to use AI to prioritize rather than simply generate more findings," said Gary Perkins, CISO at cybersecurity, compliance and managed security service provider CISO Global. "If an AI system can tell you that you have 10,000 vulnerabilities but only 12 represent a realistic path to compromise of your most important assets, that's much more useful than giving a security team another vulnerability report with 10,000 items."
Of course, giving AI agents more responsibility for vulnerability testing has raised concerns among security leaders who worry about the new risks it could create.
Claude Mandy, chief evangelist at cloud-native security vendor Zscaler, said security teams need to contain risk before granting AI agents greater autonomy.
"Unfortunately, most organizations are starting from a low foundation, lacking a contained, production-live environment that can be rebuilt easily if compromised," Mandy said. "Without the right architecture to support autonomous testing, you're conducting a live-fire exercise -- and all the associated risks -- before you even address how to constrain the agent itself to specific environments, permissions and allowed actions."
And even if organizations take those precautions, the increasing ability of AI to find vulnerabilities doesn't mean attackers will exploit every flaw discovered. Grossman also said security teams shouldn't assume that finding vulnerabilities faster means malicious actors will move more quickly to exploit those flaws.
"They're exploiting less than 1.5% of all known CVEs, of all known vulnerabilities," Grossman said. "Just because it can be hacked doesn't mean it will be. The adversaries have a business model, too."
CISOs and security teams must thus focus on vulnerabilities that pose a realistic threat rather than be overly concerned with finding more of them.
Finding that 1.5% of vulnerabilities that matter will be challenging, Grossman said, adding that organizations have "a much, much bigger haystack" to search through.
Craig Galbraith is the founder and owner of Galbraith Multimedia, an independent journalism company that provides writing, editing, video hosting, podcasting, onstage presentation and consulting services to the technology industry.
