The danger of deepfakes

The Philippines experienced the biggest jump in deepfakes among all countries in Asia-Pacific, according to a 2023 report by identity verification platform Sumsub.

It revealed that deepfakes in the region grew by an average of 1,530 percent in 2023 compared to the previous year. The Philippines saw the largest increase in deepfakes at 4,500 percent while Hongkong experienced a 1,300-percent increase, and Malaysia and Singapore, 1,000 percent and 500 percent, respectively.

Our Department of Information and Communications Technology (DICT) defines deepfakes as the mimicry of the voice or image of the person purporting to be someone else, generated by splicing photos, videos, audio, or a combination of the formats to create deceiving and false content.

Deepfakes have been described as the 21st century’s answer to Photoshopping. They are not always malicious, as when the late actor James Dean is said to be due to star in a Vietnam war movie or when a Dali Museum has a deepfake of the painter introducing his art and taking selfies with visitors. But the dangers posed by the abuse of this modern technology are real.

According to Sumsub Identify Fraud Report 2024, there has been a four-fold increase in the number of deepfakes detected worldwide from 2023 to 2024, accounting for seven percent of all fraud attempts.

DeepMedia revealed that in 2023, roughly 500,000 video and voice deepfakes were shared on social media around the world. In 2020, fake news costs the global economy $78 billion, a study by the University of Baltimore and cybersecurity firm CHEQ showed.

With the use of AI, scammers don’t have to work hard to defraud people. AI is now widely used to create videos mimicking prominent personalities in society, business, and government to convince unwitting victims to either give out sensitive information, send or transfer money, or invest in fake business opportunities.

AI deepfakes are also said to have affected election results. Last year, thousands of New Hampshire voters picked up their phones to hear what sounded like US president Biden telling Democrats not to vote in the state’s primary just days away. In Indonesia, fake videos of former president Suharto who died in 2008 endorsing the party’s candidates were posted on social media site. Soon after, Suharto’s son-in-law was elected president.

Even President Marcos became a victim when manipulated video of him urging military action against China circulated. Marcos never gave such an order. Then there was the “polvoron” video which made it appear that the President was snorting illegal drugs whose authenticity have been debunked by experts.

Recently appearing on social media feeds are deep-fake interviews by broadcasters Anthony Taberna and Jessica Soho of prominent businessmen such as San Miguel Corporation’s Ramon Ang and Gokongwei Group’s Lance Gokongwei endorsing investment opportunities with guaranteed returns.

Last year, the New York Times dubbed an AI-powered video posing as genuine footage of Elon Musk as internet’s biggest scammer. In one video taken from a shareholder meeting at Tesla, the deepfake Musk explains a product for automated trading powered by AI that can double a given investment each day.

It said that the AI was sophisticated enough that it could alter minute mouth movements to match the new script they had written for the digital fake so that to a casual viewer, the manipulation might be imperceptible.

Deloitte estimates that AI-powered deepfakes are expected to contribution to billions of dollars in fraud losses each year.

Deepfake ads have also featured prominent investor Warren Buffett and Amazon founder Jeff Bezos. Buffett said that the AI deepfake video of him was so good that he could imagine it tricking him into sending money overseas. He warned that AI scams could become the growth industry of all time.

Unfortunately, reporting them does not mean instant removal. Reporting spurious, false, sensitive, or scammy content on Facebook is a long process and they don’t always get removed, no matter how obviously wrong.

It’s mind-boggling how Meta Platforms, the parent company of Facebook, and government agencies, cannot seem to get their act together to put a stop to these scams operations and increase cyber safety and security.

The prevalence of phony ads has prompted an Australian billionaire, Andrew Forrest, whose videos were also used to create deepfake ads on Facebook, to file a civil lawsuit against Meta for negligence in how its ad business is run, claiming that Facebook’s advertising business lured innocent users into bad investments.

For its part, Meta said that it was training automated detection systems to catch fraud on its platform, but described a cat-and-mouse game where well-funded scammers constantly shifted their tactics to evade detection.

The New York Times report said that such videos cost as little as $10 to create and that scammers, mostly in India, Russia, China and Eastern Europe, cobble together the fake videos using a mix of free and cheap tools in less than 10 minutes. Shallowfakes, or those which involve simple techniques like superimposing an image onto a different background, face swapping, or changing the voice-over of real footage, are even easier and faster to make.

The DICT stressed that a comprehensive law is needed to curb the proliferation of deepfakes in the country and the country lacks specific regulations addressing the problem.

A number of bills have been filed in the 19th Congress to regulate and even criminalize the creation and distribution of deepfakes, especially the harmful ones and those used to commit crimes. Unfortunately, the House has already adjourned its session ahead of the May 2025 midterm polls and will resume on June 2 until the sine die adjournment on June 13 which means everything will have to be refiled in the next Congress.

It’s about time that Congress take deepfakes more seriously and that the President certify an anti-deepfake bill as an urgent measure before more lives and reputations are destroyed.

* * *

For comments, email at [email protected]