Apply for Exclusive Access to the 0DIN AI Security Scanner

AI systems are being deployed faster than they are being secured. Most organizations running AI in production have never stress-tested their models against the techniques real attackers use - prompt injection, jailbreaks, data extraction, and more. They are making consequential decisions on systems with unknown risk profiles.

We built the 0DIN Scanner 0DIN SCANNER Monitor, track, and analyze vulnerabilities in Generative AI and Agentic models. Explore Scanner to change that. And today, we are opening a limited, invitation-only program for organizations that want early access to run their own scans.

What the 0DIN Scanner 0DIN SCANNER Monitor, track, and analyze vulnerabilities in Generative AI and Agentic models. Explore Scanner Does

The 0DIN Scanner 0DIN SCANNER Monitor, track, and analyze vulnerabilities in Generative AI and Agentic models. Explore Scanner runs automated security assessments against AI models and systems - the same kind of testing our research team performs for enterprise engagements. It uses 0DIN's proprietary probe library alongside community tools like Garak and Microsoft PyRIT, and produces structured PDF reports that tell you:

• Your Attack Success Rate (ASR) - the percentage of adversarial techniques that actually work against your model
• Which vulnerability categories are present: prompt injection, jailbreaks, system prompt extraction, data exfiltration, and others
• How your model benchmarks against others we have tested (GPT-4o, Claude, Gemini, Meta LLaMA, and more)
• Specific hardening recommendations you can act on immediately

This is not theoretical analysis. These are the same probes and techniques documented in our public vulnerability disclosures - real attacks, real results.

The 0DIN Probe Advantage

What makes the 0DIN Scanner 0DIN SCANNER Monitor, track, and analyze vulnerabilities in Generative AI and Agentic models. Explore Scanner different is the intelligence behind it. 0DIN maintains an active research program that discovers, documents, and responsibly discloses novel AI vulnerabilities. Our current library covers 570+ probes across 719+ documented vulnerabilities.

We follow a 120-day responsible disclosure timeline. For the first 119 days after a vulnerability is discovered, it is held in a confidential, pre-disclosure state while model owners have the opportunity to address it. This means the most recent probes in our library represent threats that have not yet been made public - often techniques that attackers are already attempting in the wild.

Invitation-only access program participants get access to these exclusive, pre-disclosure probes - the leading edge of what we know about AI vulnerabilities, before it is publicly available to anyone else.

Access Tiers for Invited Participants

This is not a limited demo. Invited participants are placed into one of two tiers based on their use case and deployment scale. Both tiers include access to pre-disclosure probes not available to free users.

Small Business Tier

Built for teams actively building and deploying AI applications who need regular security validation.

• 10 scans per week with medium-priority execution
• Access to all published 0DIN probes plus the 150 oldest pre-disclosure vulnerabilities - probes that have not yet been publicly disclosed
• Community probes from Garak and Microsoft PyRIT included
• PDF reports with full findings and hardening guidance
• 1 month scan retention
• Direct access to the 0DIN team for questions and follow-up

Business Tier

Built for security practitioners and organizations with AI systems in production who need comprehensive coverage.

• 100 scans per week with high-priority execution
• Access to all published 0DIN probes plus the 300 oldest pre-disclosure vulnerabilities - twice the exclusive pre-disclosure coverage of the Small Business tier
• 0DIN Industry Variants - probes tuned for specific deployment contexts (healthcare, finance, customer service, and more)
• APA Mental Health Module - safety probes for AI systems operating in sensitive domains
• Prompt Extraction testing - assess whether your system prompt can be extracted by adversarial users
• Scheduled scans - set up recurring assessments on your model or endpoint
• Output server integration (e.g., Splunk) for security teams with existing SIEM workflows
• PDF reports with full findings and hardening guidance
• 6 months scan retention
• Direct access to the 0DIN team for questions, triage, and follow-up

Both tiers give you access to intelligence that is simply not available anywhere else. The pre-disclosure probes represent 0DIN's active research - vulnerabilities we have found, documented, and are responsibly managing through the disclosure process. If you are running AI in production, you want to know whether your system is vulnerable to these techniques before they become public.

Who We Are Looking For

We are hand-picking participants from applicants. We are looking for:

• Security practitioners responsible for AI systems in their organization
• AI product builders who are deploying or preparing to deploy models in production
• CISOs and security leadership at organizations with meaningful AI exposure
• Red teamers and AI researchers who want to benchmark their findings against an independent tool

We are specifically interested in organizations that have AI systems actively in production - not organizations still in evaluation mode. The more real your deployment, the more valuable the scan results will be.

This program is intentionally small. We are not optimizing for volume. We are optimizing for quality of engagement and quality of feedback.

How to Apply

Simply visit Apply for early access and submit your information.

We will review applications on a rolling basis and reach out directly to accepted participants. If you are selected, you will receive an invite to the platform, a brief onboarding call with our team, and full access to run your first scan.

Why Now

The window between AI deployment and AI security maturity is closing - but it has not closed yet. Organizations that understand their attack surface today will be better positioned to defend it as threats evolve. The 0DIN probe library grows every week. The sooner you know your baseline, the more useful your trend data becomes over time.

If your organization is running AI in production and you have not run a structured adversarial assessment, this is the right time to start.

---

0DIN is a Mozilla initiative focused on AI security research and responsible disclosure. Our team discovers and responsibly discloses novel AI vulnerabilities, working directly with model providers to improve the security of AI systems globally.