Wiz AI Scan Finds More Than 475 High-Severity and Critical Exposures

By Olivia Hart

More than 475 high-severity and critical exposures have surfaced through Wiz’s Scan for Good initiative, which has monitored more than 326,000 public endpoints across 17,700 domains since its September launch.

The free program pairs autonomous AI scanning with human security researchers to uncover public exposures and complex attack paths across critical infrastructure. Wiz developed it with Google DeepMind and the U.S. Cybersecurity and Infrastructure Security Agency.

Gal Nagli, head of offensive AI security at Wiz, said: “Our selection process prioritizes organizations where a successful attack could cause meaningful harm to communities and vital societal functions.” Eligible participants include registered nonprofits, open-source maintainers, educational institutions and public-health agencies that lack dedicated in-house offensive-security teams.

The initiative addresses a widening gap. Attackers increasingly use AI agents to discover zero-days and probe perimeter defenses, while under-resourced public-sector bodies, healthcare providers and open-source maintainers struggle to keep pace. Unlike large enterprises with dedicated security operations centers, many of these organizations lack visibility across their application footprints.

How the scan reached critical systems

Wiz played a leading role in uncovering the high-profile PHP2Shell memory-safety vulnerabilities that surfaced last month. That work sits alongside the program’s wider scan of public infrastructure, where exposed configurations can remain unmonitored over time.

Initial findings went beyond isolated software bugs. The discoveries included exposed administrator credentials on nationally significant archives, unprotected mobile alert channels in public hospitals, and leaked database sessions in public transportation networks.

Nagli said the exposures frequently formed multistep attack chains. Those chains combined missing access controls, exposed keys or tokens, unsafe uploads, NoSQL injections, and public CI/CD credential leaks.

The service runs through specialized AI tooling inside the Wiz Red Agent, powered by Google DeepMind frontier models, including Gemini 3.8 Flash Cyber and Gemini 4 Argon. The company says their reasoning goes beyond conventional static or dynamic code scanning to map relationships across identities, permissions, APIs and application behavior.

Nagli said the models have cyber capabilities that help defenders uncover vulnerabilities that might otherwise remain undetected. They are also trained to become better patchers and fixers, which he said helps teams fix faster.

Why discovery cannot issue the remediation order

Autonomous agents can map perimeter exposures at machine speed, but technical proof does not grant institutional authority. “Technical validation does not by itself determine institutional authority to act,” said Neil Osnato, founder of Persistence Analytics Group.

Osnato said a security team might establish that an exposed administrator credential or multihop attack path is technically real without having authority for every downstream response. Revoking a credential, altering production routing or isolating a hospital system carries operational consequences.

Those decisions require explicit human leadership, materiality assessments and clear governance frameworks, Osnato said. The more autonomous the systems become, the more important that distinction becomes. Scan for Good can locate risk and recommend action; authorized people still decide what happens in production.

What does this change for under-resourced defenders?

Scan for Good offers a practical model for using offensive AI defensively. Participating organizations receive a list of public exposures and recommendations, letting them address what attackers could exploit before an adversary acts.

For eligible nonprofits, maintainers, educational institutions and public-health agencies, the immediate benefit is free access to autonomous discovery and human validation. The corresponding duty is to triage each finding by present exploitability, potential damage and the preventive action required.

Nagli put the operating priority plainly: determine what an attacker can exploit today, what damage it could cause, and how to prevent it. For Wiz customers, he said, Red Agent pinpoints where teams should focus. Scan for Good carries that same aim into public infrastructure: surface exposure early enough to fix it before an adversary exploits it.

The program therefore does more than tally 475 exposures: it offers a blueprint for defensive AI, but one whose technical speed still depends on a slower institutional control—people empowered to validate, prioritize and fix each risk.