Critical ServiceNow AI flaw exploited days after patch release

A critical bug in the ServiceNow AI Platform was exploited in the wild, according to threat intelligence researchers Defused.

Patches for CVE-2026-6875 were released on July 13 and exploitation was first confirmed July 17.

According to NIST, the bug could enable an unauthenticated user to execute code within the ServiceNow platform. NIST and other security experts said ServiceNow users should apply the patches immediately.

“The biggest takeaway isn't just the vulnerability,” said John Strand, owner of Black Hills Information Security. “It's the difference in how quickly it was fixed. ServiceNow patched its cloud infrastructure within 24 hours, while many on-premises deployments take weeks to be fully remediated.”

Strand said that's a trend his team sees across the industry: SaaS vendors are putting the vast majority of their engineering effort into cloud platforms because that's where the business is headed and where the margins are strongest.

“Supporting on-premises deployments requires extensive regression testing across countless customer environments, and those products are increasingly receiving fewer engineering resources,” Strand said. “The result is longer patch timelines and a bigger window of exposure for organizations still running on premises infrastructure."

Denis Calderone, Principal/CTO at Suzu Labs, said the gap here repesents an example of coordinated disclosure working the right way. In this case, Calderone said Searchlight Cyber reported this to ServiceNow on April 1 and ServiceNow patched its hosted instances almost immediately.

“What took until July 13 was building and shipping a fix,” explained Calderone. “A pre-authentication sandbox-escape RCE is a hard class of bug to fix cleanly, and three months of quiet, responsible remediation before public disclosure is a feature, not a failure. The vulnerability was not sitting exposed that whole time.”

Calderone pointed out that the number that should worry people is not April to July: it’s July 13 to July 18, five days from public disclosure to in-the-wild exploitation.

“That’s the real window defenders live in now,” said Calderone. “And here’s the catch: the attackers are reaching the same code execution through a different sandbox-escape route than the published proof-of-concept, so anyone who tried to block the known technique instead of applying the patch is not actually protected. If you run the self-hosted ServiceNow, patch to the July 13 release now. Blocking the PoC is not enough.”

Roman Sannikov, global research coordinator at iCounter, said this looks like exactly the pattern his team has been flagging for a while: threat actors using AI to speed up patch reversing, so they can build a working exploit and hit organizations before they've had a chance to adopt the patch at scale.

“It's not really about whether ServiceNow took three-and-a-half months to fix the flaw,” said Sannikov. “It’s about how little time organizations now have between a patch landing and a working exploit showing up once AI is doing the reverse-engineering. A week is already fast. That window is going to keep shrinking."