While members of Congress and policymakers around the world debate how to establish guardrails for the burgeoning artificial intelligence industry, the Federal Trade Commission already has strong enforcement of algorithmic degorgement. We are introducing tools.
Also known as model removal, this enforcement strategy requires companies to remove products built on data that should not have been used in the first place. For example, if the committee finds that a company has trained a large language model on illicitly obtained data, it should remove all information along with products developed from illicitly obtained data.
Managing Director of AI Now and former FTC Advisor. “We cannot justify collecting data just to improve our products, even though it is an invasion of privacy.”
To date, the FTC has used the tool in five lawsuits against tech companies dating back to 2019, including a lawsuit against a children’s diet app and controversial data analytics firm Cambridge Analytica. there is Most recently, the commission proposed a pair of settlements asking Amazon to remove the fraudulently obtained data. The agency’s order against Amazon’s Ring invasion of privacy required the surveillance camera company to remove data products, including algorithms, from illegally screened videos. In a separate settlement with Amazon, the FTC ordered the tech giant to delete child data, location data and other voice recordings it obtained for alleged violations of the federal Child Privacy Act.
FTC officials argued the enforcement action against Amazon was a warning to other companies that may be mishandling user data in the model-building race. “Machine learning is no excuse for breaking the law,” FTC Commissioner Alvaro Bedoya said in a statement about the order. “Today’s Amazon Alexa settlement should sound a wake-up call to parents across the country, and to all AI companies scrambling to capture more and more data.”
In April, the Commission, along with the Consumer Financial Protection Administration, the Department of Justice’s Office for Civil Rights, and the U.S. Equal Employment Opportunity Commission, issued a joint statement stating, He announced plans to monitor “
FTC Chairman Rina Kern made that point again in a New York Times op-ed in May. “While these (AI) tools are new, they are not exempt from existing rules, and the FTC is vigorously enforcing the laws we are obligated to govern, even in this new market,” he said. will do,’ he said. Khan’s story in The Times followed the agency’s guidance for businesses on AI, ranging from product misrepresentation to consumer trust and deception in the first half of 2023.
One of the agency’s strengths in serving as the go-to regulator for AI is the flexibility of Congress’ powers over it. “The FTC ACT has broad applicability and was actually designed by Congress for exactly this purpose: to combat new technologies and new emerging markets,” FTC Privacy and Identity said. said Ben Wiseman, acting associate director of conservation. “Its breadth and scope will ensure that consumers are protected when these new technologies hit the market.”
When it comes to AI, he said removing models is a “key part” of the agency’s enforcement strategy. “I think the main consideration we are looking at is these AI systems, and these models are being trained on massive amounts of consumer data. What are you doing? Not just with unauthorized access, but with other disclosures that could be inadvertent or intentional?” Weisman asked.
Looking ahead, the FTC is looking at how emerging applications of AI will wipe out sensitive data as it is used in healthcare and other industries that broker sensitive data, says Weisman. said Mr. Such enforcement would be in addition to a series of recent enforcement actions related to sensitive health data by the agency.
Deleting models is not a new tool for the agency, but it has become more frequent over the past year. In May, the FTC sued an educational technology company for violating the privacy of kindergarten-age students. The settlement ordered the company to remove models and algorithms it developed using the data. Experts say disgorgement is an effective enforcement tool, not only because it imposes fines on large companies, but also because it is costly for their business models.
“It’s really about the money,” said Meredith Whittaker, chairman of the Signal Foundation and former FTC senior adviser on artificial intelligence. “They spend $5 million on t-shirts that don’t fit their staff.
A desire not to run into the FTC’s crosshairs may motivate companies to take a more cautious approach to tracking how and what data their models are siphoning. “AI systems are not designed to roll back to a specific point in time, so it’s really just complicated,” said Keegan, Cobunzweifel, managing director of the International Association of Privacy Professionals. I’m here. “It’s hard to break down a learned method over time, otherwise you may need to retrain or rethink.”
Ben Winters, a senior adviser to the Electronic Privacy Information Center Senior Advisor and head of the AI and Human Rights Project, said Amazon’s decision “hopefully motivates companies to work together.”
“One of the biggest opportunities with the use of this disgorgement is to enforce data provenance and data governance practices, because you don’t know where everything is, what kind of adjustments to what kind of data. If you don’t track what you do, enforcement agencies will be held accountable, especially not for philanthropy,” Winters said. “Rather, there seems to be an excessive set of data and algorithms that must be removed.”
The FTC’s Weisman said whether the FTC would seek remedies such as model deletion would depend on the facts of the case, but companies that use illegally obtained data would have the FTC put their data products under scrutiny and the data deletion would not be possible. He said he should know it was being considered.
Experts say there are currently some limitations to how the FTC can file lawsuits that could seek the removal of models. In both the Alexa case and two of the five cases involving the deletion of entire models, officials cited federal child privacy laws.
A federal privacy law that protects users of all ages would make it easier for government agencies to file privacy-related lawsuits. “We definitely welcome the enactment of a comprehensive privacy law,” Wiseman said. “I think that in itself makes it meaningful and makes our job easier.”
