AI Vulnerability Discovery Changes the Defense Math
Techstrong TV Interviews
•
23m
A Month That Felt Like a New Normal
AI vulnerability discovery is changing how fast flaws surface and how fast attackers can act on them. A wave of disclosures hit widely used software in a single month. PHP, Node.js and GitLab were among the names on the list. Much of that code sits quietly under the modern web, and few people think about it until it breaks.
SentinelOne Distinguished AI Research Scientist Gabriel Bernadett-Shapiro examines that shift with Mike Vizard. He compares software infrastructure to a bridge. Drivers rarely ask who inspected it last. They just assume it will hold.
Assume Attackers Have the Same Tools
Bernadett-Shapiro argues that defenders should assume adversaries already have capable AI models. Frontier providers gate their most powerful security tools behind approvals. Open-weight models, however, can be downloaded and run outside any provider's control. Restrictions can slow misuse and buy time. They cannot be the foundation of long-term defense.
He also raises a fairness question. A handful of model providers decide which defenders get early access. That leaves smaller maintainers of critical software at a disadvantage. He would like clearer rules, dependable access and funding for those maintainers.
Security, IT and Developers Need a Tighter Loop
The same models that find flaws can help validate reports, propose fixes and hunt for similar bugs. Still, experienced analysts remain essential. Asset inventory matters as much as the fix itself. Teams need to know where a vulnerable server lives before they can patch it.
Tight patching deadlines are often set by events, not regulators. The real test is how an organization weighs risk and responds.
Securing the Agent Itself
AI agents add a new insider risk. Bernadett-Shapiro expects attackers to hijack existing agents rather than deploy their own. Limiting access, such as restricting how much email an agent can read, is a sensible guardrail. In his view, securing the agent deserves more focus than securing the world around it.
He remains optimistic. AI vulnerability discovery cuts both ways, but most of the people building these tools want defenders to win.
Up Next in Techstrong TV Interviews
-
Northstar Security Gives AI Agents th...
Security teams have more tools and data than ever, yet fixing problems is still slow and manual. Alan Shimel talks with Amer Deeba, CEO and co-founder of Northstar Security, about the launch of his new startup and its plan to give AI agents the context they need to fix security risks.
Why No...
-
Splunk Finds Most Teams Cannot Contro...
More than half of IT teams say they cannot accurately control their AI costs. Alan Shimel talks with Greg Leffler, director of developer evangelism at Splunk, about the 2026 Splunk State of Observability report and what it reveals about AI, predictability and the future of operations.
From t...
-
IT Is Becoming an HR Department for A...
NVIDIA CEO Jensen Huang once said the IT department will become the HR department for AI agents. Alan Shimel talks with John Capobianco, head of AI and developer relations at Itential, about what managing AI agents looks like in practice for network and infrastructure teams.
From the factory...