Bumble wants AI to stop toxic relationships before they start

Every dater knows the dread of matching with someone charming, only to realize weeks later that you have walked straight into a manipulative nightmare.

Bumble thinks artificial intelligence might be able to spot those red flags before you ever say hello.

Speaking on the Leaders in Innovation podcast, Bumble founder Whitney Wolfe Herd outlined a bold vision for the platform: using AI not just to find better romantic matches, but to actually prevent toxic relationships before they happen.

The timing comes amid major leadership shifts. On November 6, Bumble announced that Wolfe Herd will step down as CEO, handing the reins to current Slack CEO Lidiane Jones on January 2. Wolfe Herd will transition to executive chair, staying heavily involved in the company's 10-year roadmap, public policy, and social impact.

Designing algorithms through the feminine gaze

When Wolfe Herd founded Bumble in 2014, the idea of women making the first move was countercultural. She built the app around a blunt observation: the internet was never engineered for women, and the 2010s mobile boom mostly bypassed their lived safety concerns.

Now, she sees history threatening to repeat itself in AI. If models are built on datasets, prompts, and engineering teams that skew overwhelmingly male, women will once again bear the brunt of digital harm.

Wolfe Herd argues that female scientists, doctors, and gender studies experts need direct control over how these systems are trained. Bumble already has some proof of concept: back in 2019, it introduced an AI-powered lewd photo detector boasting a 99% accuracy rate, which later helped spur state-level legislation against cyberflashing.

Taking that logic to romantic pairings is the next frontier, though Wolfe Herd did not reveal the technical mechanics behind predicting relationship toxicity.

Why it matters

Swiping fatigue is real, and toxic interactions are the fastest way to drive users off dating apps forever. If Bumble can genuinely protect daters from bad actors and emotional burnout, it gives people a compelling reason to stick around.

Yet we think this ambition walks an extraordinarily thin line. Toxicity is deeply contextual and subjective. To predict whether two people will become emotionally harmful to one another, an algorithm would likely need access to deeply personal data and chat dynamics. What starts as a protective shield could quickly look like intrusive surveillance.

From an industry standpoint, bringing in enterprise veteran Lidiane Jones while refocusing Wolfe Herd on policy and AI philosophy is a clever division of labor. Dating apps desperately need a second act beyond the swipe mechanics of the last decade, and leaning into AI-driven safety is Bumble’s strongest differentiator. But turning human heartbreak into a solvable data problem might be a much taller order than filtering out bad photos.