SiliconANGLE theCUBE · Nov 5, 2026 · 20m · 12m saved
With Juan Loaiza, Executive Vice President of Mission-Critical Database Technologies at Oracle
In March 2026 Oracle ran an early-access AI model against its own software and found dozens to hundreds of security vulnerabilities nobody knew were there. Juan Loaiza, executive vice president of mission-critical database technologies at Oracle, told Dave Vellante on SiliconANGLE theCUBE that the company treated it as an emergency, not as routine maintenance.
Most companies still treat AI security as a job for the network perimeter. Juan's answer was to push the rules about who can see what all the way down into the database.
"This is not a normal event we can just kind of schedule anytime we want. I mean, this thing is finding dozens to hundreds of security vulnerabilities. Our number one job has to be to fix it." — Juan Loaiza
Juan oversees Oracle's mission-critical database technologies, and his team had started building defenses against AI-driven threats several years before the March findings.
The full interview is covered here so you can skip it. 20 minutes of audio, 8 minutes of reading.
Here are the 9 takeaways that matter.
Key Takeaways
An early-access AI model found dozens to hundreds of unknown vulnerabilities in Oracle's code
Oracle's quarterly patches have grown much larger since March because of the volume of fixes
No one can review tens of thousands of lines of AI-written code, so the database has to enforce access itself
Oracle's deep data security, which began shipping in April 2026, passes each user's identity down to the data
A wrong AI answer keeps coming back until its root cause is fixed, Juan says
Formal ontologies were built because computers were dumb, and plain English now does the job
Some customers still think a firewall is enough
1. A Code-Red Moment
Juan said the trigger came in March, when Oracle got early access to a frontier AI model he called Mythos and pointed it at its own code.
An early-access AI model began finding Oracle's vulnerabilities immediately
"Back in March, we got early access to the Mythos model, and we ran it against our code, and it started finding all these security vulnerabilities." — Juan Loaiza
He told his team to treat the findings as an emergency. Better that Oracle finds the flaws, he said, than attackers using the same tools.
Oracle's database chief called the findings a code-red moment
"This is a code red moment. The house is on fire. We got to get on top of this thing because we have to be able to protect our customers from all these issues." — Juan Loaiza
