FTC Warns of 'Quiet' TOS Changes for AI Training: Here’s Why This May Not Be Sufficient.

This week, the FTC issued a warning to companies about the risks of subtly altering their Terms of Service (TOS) to exploit user data for AI training. The agency cautioned that such actions could be deemed unfair and deceptive practices.

“Companies might be tempted to change their privacy policies to expand their use of customer data,” noted the FTC in a blog post. “To avoid potential backlash over privacy concerns, these changes may occur without clear notification. However, businesses should be aware that violating user privacy commitments could lead to legal repercussions.”

Although the FTC's warning is significant, it comes after notable industry shifts. In July 2023, Google modified its privacy policy to allow the use of public data for AI model improvement. Similarly, in August 2023, reports emerged that Zoom had quietly updated its TOS to permit AI training on user data without an opt-out option.

In June 2023, frustrations arose among Adobe Stock creators when they discovered that Adobe was using their uploaded work for its Firefly AI model without prior notification. “I don’t recall receiving any email or notification about changes to the terms of service,” one UK-based creator remarked.

The reality is that most users do not read or fully comprehend the fine print of Terms of Service documents. A 2019 study analyzing the TOS of 500 popular U.S. websites, including Google and Facebook, revealed that over 99% of these documents exceeded the level that most adults typically read. Moreover, a 2017 Deloitte survey indicated that 91% of U.S. consumers consent to legal terms without reading them, with 97% of those aged 18-34 doing the same.

Bradley Kuhn, a policy fellow at the Software Freedom Conservancy, emphasized the importance of reading TOS during an October 2023 FTC roundtable. “If there’s one takeaway, it’s to carefully read the terms and conditions. You may have inadvertently agreed to include your creative works in a company’s machine learning datasets. It may take time, but it's the only way to understand what rights you may have given up.”

Interestingly, many recent TOS updates simply clarify existing clauses. According to Katie Gardner, a partner at Gunderson Dettmer law firm, updates typically reflect evolving practices, and regulations like the CCPA mandate annual privacy policy revisions. “Companies are obligated to notify users of significant changes for them to be legally enforceable,” she explained. “In Zoom's case, changes might seem less material because they only reaffirmed existing rights.”

Furthermore, Gardner noted that many updates are in response to regulatory scrutiny. “The methods companies use to obtain consent for data usage in AI training are under increased review,” she mentioned, pointing to the FTC's actions against Ring and Amazon regarding transparency in user data usage.

Users can expect additional TOS surprises in the coming months. Recently, planned updates by Findaway Voices, a Spotify audiobook platform for independent authors, raised concerns about audio narration being repurposed for AI training. In response to online feedback, Spotify acknowledged the confusion and is working on clarifications before the updates take effect on March 15, 2024.