

Google plans to automatically publish qualifying reports from its Google & Alphabet Vulnerability Reward Program, widening access to examples that security researchers and AI-assisted workflows can learn from. Announced on November 19, the policy applies to eligible submissions from December 1, 2026; Google expects disclosed reports to begin appearing in early 2027.
The official announcement describes a response to a changing reporting environment. Google says duplicate reports have increased fifteenfold compared with 2023, while average report length has tripled. These are Google's observations, not an independent measurement of how much AI alone caused either change.
The clocks start at different points
Automatic publication requires a reward of at least $10,000, a submission on or after December 1, at least thirty days since the issue was fixed, and at least seven days since both the reward and researcher notification. Special events or grants involving unreleased products or features are excluded. This is not a policy of immediately exposing every newly reported vulnerability.
Researchers will be able to request anonymous disclosure and propose edits before publication. Google also describes its own privacy-preserving edits, including removal of sensitive identifiers and cookies. Readers should consult the linked program rules for the applicable terms rather than treating a news summary as a disclosure agreement.
More examples may help; more text does not necessarily help
Our analysis: a growing pile of reports is useful only if reviewers can distinguish reproducible findings from repeated or unsupported claims. A public record of fixed, rewarded work could offer a more concrete teaching resource than a stream of confident vulnerability allegations. That value depends on the quality of the evidence and on preserving enough context to understand the original failure.
The timing rules also matter when building research datasets. A report's submission, fix, reward and public disclosure dates describe different stages. Combining them into a single discovery date can produce misleading timelines, particularly when comparing human work with agent-assisted research.
Publication will not by itself demonstrate that an AI tool can find an unfamiliar flaw or safely handle an undisclosed one. Future evaluations should account for whether the system already saw the disclosed report. Otherwise an apparent security breakthrough may partly reflect recognition of public material.
For now, the news is a forthcoming transparency policy with explicit conditions. We have not inspected a new disclosure collection or tested an AI bug-finding system against it. The first published reports will be the point at which researchers can assess how useful this change becomes in practice.
Original source-reviewed reporting and clearly identified editorial analysis. No independent model evaluation, installation, scan or cloud deployment was performed.






