The Dangers of Audio Deepfakes, RD x Comcast

Deepfake of the Week

Video and image deepfakes are becoming increasingly more prominent online, especially as political campaigns start using them in misleading ways. (We discussed this in last week’s newsletter, as well as in a New York Times cover story on Monday.)

Yet while AI-generated visual manipulations and fabrications run rampant in the digital realm, audio and voice deepfakes are wreaking havoc in the real world. With cheap off-the-shelf technology, anyone can use seconds of recorded audio to fake a kidnapping or deceive family members with relative ease. In a connected world where voiceprints verify your identity at your bank and audio messaging sees hockey stick growth amongst all generations, the potential for causing lasting, tangible harm is only growing with each new advancement.

Reality Defender is decidedly not a deepfake creation company. Yet to be a truly best-in-class deepfake detection platform, our team stress tests our technology against every conceivable new method and app (and thensome), becoming experts in deepfake creation along the way. After seeing new consumer-oriented apps appear that require only a few seconds of sound to clone someone’s voice with nearly note-perfect accuracy, we put said apps and our platform to the test. 

How easy is it to deepfake someone’s voice?

Click the button to hear a deepfake of Reality Defender Co-Founder and CEO Ben Colman explain just how easy it was to clone his voice.

Listen

What does Reality Defender’s platform say?

Reality Defender Joins Comcast NBCUniversal’s Lift Labs

The Reality Defender team is excited to announce our participation in the 2023 Comcast NBCUniversal LIFT Labs Generative AI Accelerator.

Over the last six weeks, our team collaborated with key figures from Comcast and NBCUniversal to herald a new phase of Reality Defender's premier deepfake detection technology. We look forward to sharing more information on this work in the near future.

NIST to Address AI Risks

The U.S. Commerce Department's National Institute of Standards and Technology (NIST) is launching a public working group to explore the potential and risks of generative artificial intelligence (AI). Supported by President Biden, this launch comes as regulators worldwide scramble to form regulations for the rapidly advancing AI-driven technology.

A New AI Regulation Framework in the U.S.

US Senate Majority Leader Chuck Schumer introduced a framework for regulating artificial intelligence, pushing for Congress to get a better understanding of technology with a focus on security, accountability, and innovation. In the hours after its introduction, the framework received mixed reviews, with some experts expressing concern that the proposed "insight forums" for lawmakers could delay the regulation of AI, while others view it as a necessary step to stimulate discussion around the topic.

U.K. Cracks Down on Deepfake Porn

Amendments to an online safety bill in England and Wales will criminalize the sharing of explicit deepfaked images without consent and ease the burden of proof for charging perpetrators of revenge porn. The legislation will also introduce a specific offense for threatening to share intimate images, with potential sentences up to two years for intended harm and inclusion on the sex offender register for offenses committed explicitly for sexual gratification.

Editor’s Note: No Newsletter Next Week

The Reality Defender newsletter will take a brief pause due to the coming Fourth of July holiday and return on July 11th.

Deepfake News

• AI-generated child sex images are proliferating on the dark web, making it harder for authorities to combat child exploitation. (Ars Technica)
• The Information has a feature on Adobe’s approach to AI and their Firefly enhancements in existing products. (The Information)
• Aloud, a company in Google’s Area 120 incubator, is bringing AI-powered multilanguage dubbing to YouTube. (Aloud)
• A new Midjourney update was revealed and features new photorealistic flourishes. (Ars Technica)
• Voicebox, a new generative AI model from Meta, can perform high-quality audio editing and generation tasks (Meta)

AI News

• Time reveals OpenAI lobbied for changes to the EU's AI Act to reduce regulations on its general purpose AI systems, despite their potential use in high-risk scenarios. (Time)
• Wired has a spotlight on Pause AI, which is warning against rampant AI development that could potentially cause human extinction. (Wired)
• Congress and all related staff members, should they want to use ChatGPT, must use ChatGPT plus due to its privacy features according to a new memo. (The Verge)
• Google’s DeepMind is working on a chatbot that will allegedly outwork ChatGPT. (TechCrunch)
• Newark schools are using a new AI-powered teaching aid from Khan Academy — and finding it lacking. (The New York Times)

AI in Entertainment News 

• Sony Music recently announced and then hired for a new EVP of AI role. (Billboard)
• The opening credits sequence of Disney's latest Marvel show, Secret Invasion, was revealed to be created using AI. (Vulture)
• As it turns out, that new Beatles song may not be so artificially created after all. (Variety)
• Deepfakes are having a material impact on K-Pop. (DNAIndia)
• Live 95.5, a Portland radio station, now has the “first” AI-powered DJ. (Twitter)
• Director Christopher Nolan is apparently not surprised by the warnings of various AI-related dangers. (Variety)