Roblox safety VP Eliza Jacobs says “ticking a box” for age is no longer enough

Roblox's safety play is getting a technical upgrade, and its top policy voice is blunt about why. Eliza Jacobs, Roblox's vice president of safety product policy, told NBC News that “ticking a box” to claim you are 13 or older is “not enough anymore.” In the same conversation, Jacobs said Roblox is “optimistic” its new facial age estimation tech will “continue to get better,” and NBC invited kids to try the process.

That “video selfie age check” is the centerpiece. NBC brought a group of kids to test Roblox's new method for verifying age, and they were not able to get around it with a fake mustache. Jacobs claims Roblox's facial age estimates can typically land “within 1.4 years of a child's exact age.” In other words, Roblox is not just collecting a checkbox answer, it is using computer vision to estimate age, then using that estimate to decide how a user is bucketed.

Roblox announced plans in April to begin using its video selfie age estimation tech to sort players into age brackets. Users under 16 and unde... (the source text is truncated), but the direction is clear: the platform is moving from a self-reported threshold toward a measurement-based approach to age gating. For executives, that is a meaningful operational shift. It creates a new dependency on model accuracy, user experience, fraud resistance, and ongoing system tuning. It also changes the burden of proof. If something goes wrong, a “we asked nicely” approach becomes a “we implemented a system” approach, and regulators or plaintiffs can evaluate whether the system was adequate.

This kind of change sits right in the crosshairs of modern platform risk. Age verification is one of those requirements that looks simple on paper until someone tries to game it. A checkbox is cheap and fast, which is exactly why it fails under pressure. The NBC testing detail matters because it shows the friction Roblox is attempting to introduce. If a common trick, like a fake mustache, does not work, that suggests Roblox's checks are more than a single visual filter. It is an estimation process, and Roblox is betting that it will get better over time.

Jacobs' “within 1.4 years” claim is also the kind of metric decision-makers should treat as a strategic input, not just PR. If the estimate can land within that window, Roblox can set age bracket thresholds with a margin for error. But margins also create corner cases. A system that estimates within a range will inevitably misclassify some users near boundaries, especially if lighting, camera quality, or user behavior affects inputs. Even when the tech is improving, accuracy is never uniform across demographics, devices, or environments. That is the hidden governance work executives should expect: monitoring error rates, handling disputes, and designing a fallback path that does not just punish kids who get flagged.

There is also a product implication. Age checks are not neutral. They can slow onboarding, add steps to play, and potentially create a perception of distrust among families. In a world where engagement is the product, any additional friction can affect growth. Roblox is essentially balancing two competing incentives: protecting children and complying with age-related expectations, while minimizing drop-off from the verification flow. The reason Jacobs is saying the checkbox is “not enough anymore” is that platform safety teams are likely being held to higher standards than they were in the era of self-attestation.

Second-order implications extend beyond Roblox. If Roblox is making this move, other platforms that rely on self-reported age thresholds face a reputational and regulatory risk. The benchmark shifts from “did you ask” to “did you verify in a way that resists obvious attempts to bypass.” In board terms, the safety and policy function is moving closer to engineering reality, and oversight has to follow. Boards and executives should expect more scrutiny on how models are trained, how updates are rolled out, how performance is measured, and what happens when estimates are wrong.

Roblox is also signaling confidence in iterative improvement. Jacobs told NBC News that Roblox is “optimistic” the facial age estimation tech will “continue to get better.” For decision-makers, that phrase is not just optimism, it is an operating stance. It implies ongoing refinement, which means budgets and accountability for continuous model governance. The strategic stake is straightforward: platforms that treat age verification as a one-time checkbox are likely to look increasingly outdated. Platforms that treat it as an evolving system will still have problems, but they will at least be able to show they are upgrading safeguards in response to real-world bypass attempts.

The moment here is not just a new feature. It is Roblox redefining what “good enough” looks like in age gating. And if “ticking a box” is no longer enough, the companies that will win are the ones that can prove their safeguards actually work, in practice, not just in policy language.