As offensive AI capabilities improve, empowering defenders has become a global imperative. The damage cyberattacks can inflict upon under-resourced organizations and the communities that depend on them is an issue that affects us all. It is in this spirit that we are launching Scan for Good, an initiative that uses AI - coupled with human security researchers - to uncover public exposures and complex attack paths across public services, critical infrastructure, and nonprofits.
Scan for Good helps defenders secure vital systems before malicious actors can exploit them. Early results prove its impact. Our efforts spent validating findings and collaborating with the affected organizations on remediation have uncovered hundreds of public exposures that were fixed as a result of Scan for Good. Later in this post, we will share some of these stories.
We would like to thank the team at Google DeepMind for partnering with us to ensure the success of Scan for Good. We’d also like to thank the Cybersecurity and Infrastructure Security Agency (CISA) for engaging with us on the project to offer their collaboration and guidance.
At a time of evolving threats, defensive vulnerability discovery helps strengthen the nation’s digital infrastructure. Under President Trump’s Executive Order 14409, CISA promotes the lawful and responsible adoption of AI to accelerate vulnerability discovery and strengthen cybersecurity resilience across government, industry, and the essential organizations that support our communities. Together, we can build a resilient ecosystem in which those who protect our society have access to advanced technology and leading cybersecurity expertise.
Nick Andersen, Acting Director, CISA
AI is changing exploitabilityAs AI makes rapid advances, much attention has been given to models discovering new zero-day vulnerabilities, which they do largely by reasoning over and analyzing code. Through our work building the Wiz Red Agent – an AI-powered, context-aware pentester – we saw another gap, with more immediate implications: the security of applications exposed and reachable on the public internet.
That is where we chose to focus this project.
In real-world environments, critical risk often does not come from a single vulnerability in code. It can emerge from the combination of configurations, permissions, identities, APIs, and application behaviors that may appear benign individually but create dangerous attack paths when connected.
Historically, finding and understanding these paths at scale required significant time, expertise, and manual investigation. AI is changing that, making it dramatically easier to discover exposed systems, understand how they behave, and connect weaknesses into real attack paths.
We have already seen this shift firsthand. The Red Agent has identified thousands of high and critical exposures in production environments that could put organizations at immediate risk. As these capabilities become more accessible and widespread, security teams will need new tools, expertise, and continuous research to keep pace … resources that not every organization has.
Scan for Good brings these capabilities to organizations where successful exploitation could have an outsized impact, including critical infrastructure, public services, healthcare, and nonprofits.
Where authorized, we will use the Wiz Red Agent and additional internal AI research capabilities to examine publicly facing websites, APIs, and applications (see below for more details on testing scope and disclosure). This work will help organizations find and fix critical exposures before attackers do, allowing them to stay ahead of adversaries as AI capabilities advance.
AI models like Gemini Cyber offer us the opportunity to advantage defenders over attackers. This is especially needed for organizations that serve vital functions in society, but many do not have the resources they need to protect against attacks. Scan for Good rises to this challenge by leveraging AI to uncover risks and protect critical infrastructure, non profits, and other public-interest organizations. In the coming months we will work to scale that impact globally.
Raluca Ada Popa, Head of Gemini Security @ Google DeepMind
What we’ve learned so farBelow are some examples of ways in which Scan for Good is already helping public-interest organizations and major technology providers find serious internet-facing risks before attackers do. Every example was discovered autonomously by the AI systems powering the program, safely validated by Wiz Research only far enough to confirm real-world impact, and privately disclosed so the affected organization could remediate it.
Public-interest organizationsThese examples illustrate early results regarding one of Scan for Good’s stated objectives: to help protect organizations whose security is critical to the communities they serve.
In each of the cases below, we partnered with the affected organization on remediation efforts.
National archive: An administrator key was exposed on a public web server, enabling read, write, and delete access to 8.8 million files in a nationally significant archive of an EMEA country. We helped remediate the exposure by assigning the correct set of permissions.
Public hospital: Missing access controls exposed staff contact information and gave anyone online control of a hospital-wide mobile alert channel.
Private hospital: An unsafe upload on a public appointment-booking site gave control of a hospital server and exposed patient identifiers, clinical information, and consent signatures.
Municipality: A public data service exposed sensitive personal, health, and financial information for roughly 5,000 elderly residents. We confirmed the risk without collecting a bulk dataset.
Public rail operator: A leaked production database exposed active administrator sessions, enabling control of routes, schedules, service announcements, and administrator accounts. We helped the operator secure the system before public transportation could be disrupted.
Foundational digital infrastructure In addition to under-resourced and public sector organizations, Scan for Good also focuses on essential tech platforms that provide the digital backbone of society. Below are several examples; as with the previous, we partnered with the affected company on remediation efforts following disclosure.
Leading AI training-data platform: Missing access controls utilizing a NoSQL Injection created a path to leak and alter customers’ proprietary AI-training data and projects.
Website-building and commerce platform: A zero-day flaw in a shared payments service exposed customer names, card brands, expiration dates, and partial card numbers across multiple stores. We confirmed it was a platform-wide issue.
Enterprise data and AI platform: Vulnerable public CI/CD workflows exposed credentials for an internal issue tracker and production marketing database, putting internal tickets, proprietary data, and customer records at risk. We worked with the provider to secure the workflows, rotate the credentials, and remediate downstream exposure. Read the first part of the published case study.
Cloud infrastructure provider: A credential exposed in public website code could have been used to publish malicious software across over 500 production container images supporting a flagship AI service. We proved the reach without changing an image and worked with the organization on containment.
Across these cases, a small public signal – be it a forgotten route, a missing permission check, or an exposed credential – quickly and autonomously escalated access to sensitive data, infrastructure, or administrative control. Scan for Good uses that speed for defense: find the path early, prove only the impact needed, and help close before it becomes an incident, working in collaboration with the affected teams.
This work builds on earlier Wiz Research into Base44, DeepSeek, Moltbook and Snowflake. In these instances, simple public exposure led to significant impact. Scan for Good applies those lessons more broadly and, after remediation and coordinated disclosure, shares the patterns anonymously so defenders can adapt.
Acting before an attack: Commitment to proactive security Our goal is simple: find serious risks, help organizations fix them, and do so before those risks are exploited.
We will prioritize infrastructure where a successful attack could cause meaningful harm, including public services, critical infrastructure, nonprofits, open-source projects, and under-resourced organizations.
We will only conduct testing where it is authorized, either where the organization has an authorized bug bounty program or established vulnerability disclosure policy, or with explicit authorization. Any organization concerned about an exposed service can apply for a Scan for Good assessment.
Every potential finding will be reviewed and validated by a human researcher. AI can help us investigate and explore more systems and possibilities, but humans will remain responsible for confirming impact and making disclosure decisions.
When we identify a serious issue, we will contact the affected organization privately, provide clear technical information, and work with the team to support remediation where appropriate.
Scaling Scan for Good with Gemini 3.8 Flash CyberScan for Good uses advanced cyber models for automated defense, with a goal of creating ecosystem-scale impact. In this sense, it shares some common DNA with Google DeepMind’s Fairwind Program (Fairwind was designed as a way to safely provide defenders early access to powerful frontier capabilities).
Our partners at Google DeepMind were critical to both initiatives: the AI-powered tooling behind Scan for Good is powered by the Gemini family of models, especially the new Gemini 3.8 Flash Cyber. Through our collaboration with the Google DeepMind team, Gemini 3.8 Flash Cyber’s frontier cybersecurity performance is helping uncover complex attack paths in foundational AI and cloud technology, and fueling Wiz as we work with public services, nonprofits, and other critical infrastructure providers to ensure they stay ahead of AI-powered adversaries.
In the coming months, we’ll release more details on how Google DeepMind is helping us continuously improve our Red Agent use case. We also look forward to sharing real-world examples of how Gemini 3.8 Flash Cyber has had an impact on organizations around the world through Scan for Good.
Turning AI insight into collective knowledgeScan for Good is not only about fixing exposures. It also aims to help the broader security community understand how AI is changing exploitability, and which weaknesses are truly meaningful.
After the findings we report have been remediated, we plan to publish anonymized research detailing the underlying vulnerability patterns, the impact AI had on practical exploitability, and how organizations can put this into practice in scanning their own environments.
We will focus on the lessons rather than the affected organization.
We hope this work gives defenders a practical view of what offensive AI can do today, where traditional approaches may fall short, and how exposure-management programs need to evolve.
Responsible researchScan for Good is built around authorization, minimal and non-destructive validation, human oversight, private disclosure, and clear stopping points. We will minimize interaction with live systems, avoid unnecessary access to sensitive information where possible, and give organizations a reasonable opportunity to remediate.
We will not treat model-generated hypotheses as vulnerabilities. Findings will be validated by experienced researchers, and deeper testing will only happen where it is authorized.
Scan for Good is our effort to turn advances in AI into an advantage for defenders: finding critical exposures before attackers do, helping organizations remediate them, and sharing what we learn so the broader security community can adapt.
Organizations concerned about exposed infrastructure can apply for a Scan for Good assessment at wiz.io/scan-for-good/apply.