Public Disclosure Decisions
What you'll learn: You'll understand the key factors that justify making AI or ML vulnerabilities public, even when typical disclosure processes haven't resolved the issue.
When Private Disclosure Isn't Enough
You've already learned about responsible disclosure channels and the 90-day standard. But sometimes the normal path breaks down. Public disclosure means revealing a vulnerability openly—through blog posts, conference talks, social media, or research papers—before it's fully patched or acknowledged.
This isn't a decision to take lightly. It can harm users still exposed to the risk, damage vendor relationships, and create controversy. But in certain situations, it becomes ethically necessary.
Three Critical Factors for Going Public
1. Unresponsive Vendors
If you've reported a serious vulnerability—like a prompt injection that leaks private data, or a model inversion attack exposing training data—and the vendor ignores repeated contact or refuses to acknowledge the issue past the 90-day window, public disclosure may be the only way to protect users and create accountability.
2. Active Exploitation
When you observe a vulnerability being actively exploited in the wild—for example, attackers using a discovered jailbreak to generate harmful content at scale—waiting quietly becomes irresponsible. Users and defenders need to know now to protect themselves, even if the fix isn't ready.
3. Public Interest
Some vulnerabilities have broad societal implications: a facial recognition system misidentifying people of certain ethnicities, or an LLM leaking personally identifiable information. When the impact extends beyond individual users to affect civil rights, democracy, or public safety, transparency serves the greater good.
Balancing Harm and Benefit
Think of it like a fire alarm. You don't pull it for minor issues—but when the building is burning and officials aren't responding, you act to save lives, even if it causes panic.
Key Takeaway: Public disclosure is justified when vendors are unresponsive beyond reasonable timelines, vulnerabilities are being actively exploited, or the issue carries significant public interest—always weigh potential user harm against the harm of continued silence.