As AI becomes more sophisticated and easier to use to mimic real people, businesses face new risks to their reputation and bottom line — and women founders are especially exposed.

Already, the vast majority of deepfakes target women, and Americans are worried about how much of a threat they pose, with 67% saying they are concerned about AI being used to exploit women . For women business owners who already face a myriad of challenges in getting to the top, deepfakes can be weaponized for extortion or damage the reputation they have built with their audience.

Alisha Outridge, founder of Byte&Chord , a venture studio building AI products, sees a clear throughline between the threats women leaders have long faced, including sexual exploitation, and the weaponization of AI against businesses. “I’m a public-facing female executive in tech. I've had stalkers, and I've had companies approach me, offering a revenue share to license a clone of me. So I know the demand is already there,” she says. “My concern is what happens when those stop being offers. As my presence online grows, so does the training data, and I expect this to follow the same arc as revenge porn: something done to you retroactively, without consent, with remedies that arrive years too late.”

AI quickly reproducing a founder’s identity is what especially worries women founders because deepfakes make it increasingly difficult for customers to distinguish between the person they trust and an AI fake version of them.

“A deepfake impersonating me could undo years of that trust in a single viral moment,” said Maggie Olson, founder and CEO of Nova Chief of Staff , an education and professional development platform.

“What concerns me most is how fast misinformation moves and how slow correction travels. For a bootstrapped company where my reputation is the product, that is not a recoverable situation overnight,” said Olson, who also previously served as Chief of Staff at T-Mobile.

Natanya Wachtel, founder of New Solutions Network , says she knows “from the inside” how convincing the technology already is.

“A deepfake borrows the trust an audience spent years building with you, and for a founder, that trust is the whole business, and right now it can be copied faster than it can be defended. For a small business, the founder is often the brand. The more of you that’s public, the more there is to copy,” said Wachtel

The consequences can be expensive. In the past couple of years, deepfakes have already tricked large firms like Arup into wiring $25 million to scammers . In Arup’s case, the AI being used to impersonate the CFO only had to fool one employee at the right moment.

Business owners targeted by deepfakes have some legal avenues to protect their reputation and businesses. If a deepfake inflicts commercial damage or misleads consumers, businesses can pursue civil claims for defamation or false light, alongside federal trademark suits under the Lanham Act. Furthermore, business owners can demand immediate, 48-hour content removal from online platforms by leveraging the federal TAKE IT DOWN Act for non-consensual intimate deepfakes. States also have laws against identity theft and criminal harassment.

These laws, however, are poorly enforced, because they pass without the funding, training, or staff to make them work — and they primarily focus on combating harms after the damage is done. Current legal avenues are also riddled with loopholes, as Big Tech fights them in court , claiming that some laws infringe on free speech.

If, as the U.S. government recognizes, small businesses are “ the backbone of the U.S. economy ,” then we must go further in regulating AI systems so they can’t produce these non-consensual deepfakes for harmful purposes in the first place.

Disclosure: Women Who Tech and RAD Campaign, which I founded, commissioned the U.S. poll of 1,015 adults age 18 and older cited in this article. The results were weighted to ensure proportional responses. The Bayesian confidence interval for 1,000 interviews is 3.5, roughly equivalent to a margin of error of ±3.1 at the 95% confidence level.