'Telltale signs' of AI videos are becoming undetectable — and more dangerous

MANILA, Philippines — Content generated by artificial intelligence may soon be impossible to detect based on visual cues alone, increasing its potential to be used in disinformation campaigns, experts warn.

Carlos Nazareno, digital artist and director for rights at advocacy group Democracy.Net.PH, said generative AI models are improving rapidly in mimicking reality, making it harder to spot visual inconsistencies.

“Whatever telltale signs we use to spot [generative] AI videos are going to be obsolete in months or years,” he told Philstar.com in an interview. 

“I don't know how easy it will be to tell if videos are AI or not because of the rapid pace [of their development],” Nazareno added.

He noted that past red flags, such as distorted body features like six fingers, are now being corrected as models evolve.

Hyperreal visual, viral fakes. These advancements are already evident in tools like Google Veo 3, the latest version of Google’s AI video generator released in May. The update introduced the generation of hyperrealistic visuals and synchronized audio.

The introduction of Veo 3 fueled a recent wave of viral AI-generated videos across social media platforms like TikTok and Facebook. Many of these simulate news segments, such as those that appear to feature Filipino reporters conducting man-on-the-street interviews.

One of the most controversial examples depicts two students disagreeing with the impeachment of Vice President Sara Duterte — a video Sen. Bato dela Rosa shared on his official Facebook page on June 15.

RELATED: Dela Rosa draws flak over AI video of students opposing Sara Duterte's impeachment

While fact-checkers flagged oddities such as garbled signage and non-existent school logos, street interviews conducted by Philstar.com indicated that many college students—considered “digital natives” — failed to recognize the video was fake. 

“Hindi ko malalaman na AI ‘yun at first glance. Kapag paulit-ulit tignan, siguro may mapapansin ako,” a student from the University of the East said. “Kung magshe-share [ka] ng ganoon, kailangan siguro muna i-investigate lalo na kung hindi ‘yun galing sa credible na source.” 

(I wouldn’t be able to tell it was AI at first glance. Maybe if I watched it over and over, I might notice something. If you're going to share something like that, you should probably investigate it first, especially if it didn’t come from a credible source.)

Tools for political manipulation

Nikko Balbedina, fact-checker and multimedia editor of PressONE.PH, warned that generative AI poses serious threats when weaponized in political influence operations, which are deliberate efforts led by government or private entities to manipulate public opinion.

“It is interfering with our shared understanding of reality. Hindi na ganun kahirap magsinungaling kasi ngayon kayang-kaya na natin magmanufacture ng ebidensya. It becomes dangerous lalo na sa ganitong sitwasyon na pinag-uusapan [ang] impeachment ng isang opisyal ng pamahalaan,” he told Philstar.com in an interview.

(It is interfering with our shared understanding of reality. It’s no longer that difficult to lie because now, it’s so easy to manufacture evidence. It becomes dangerous, especially in situations like this, when we’re talking about the impeachment of a government official.)

He said the continuous spread of AI-generated disinformation may erode public trust in the democratic process of impeachment and deepen political divides.

Several such instances were observed the last midterm elections, including AI-manipulated videos of a foreigner allegedly endorsing Kingdom of Jesus Christ founder Apollo Quiboloy’s senatorial run and First Lady Liza Araneta-Marcos supposedly campaigning for former President Rodrigo Duterte’s DuterTen slate.

Following Dela Rosa’s repost, the Cybercrime Investigation and Coordinating Center urged the labeling of AI-generated content. But Balbedina argued that the burden of labelling content should not fall solely on users.

“These tech giants, they have the capacity to make their own software that can detect AI,” he said, noting that malicious actors will likely avoid labeling deceptive posts.

Focus on message, not just visuals. Nazareno of Democracy.Net.PH urged social media users to pay closer attention to the content's messaging rather than relying on visual or audio cues. 

He explained that sensational or emotionally charged content could be designed to mislead. Users should cross-check posts with credible news and fact-checking sources.

“It’s also difficult because we're living in a post-truth world wherein actual news headlines read like satire because of the actions of some leaders. So we really have to look at these things with a critical eye and mind. That’s our best defense,” Nazareno said.

Gaps in education, policy 

To keep pace with the rise of AI-generated media, Nazareno stressed the importance of robust media and information literacy programs, or MIL, in the Philippines.

While the Department of Education (DepEd) currently includes MIL in senior high school curriculum, it does not cover topics like AI, disinformation, or fact-checking. 

While MIL is offered now, it may no longer be taught as a standalone subject. DepEd’s revised curriculum for senior high school released in May reduces core subjects from 15 to five. One of the new subjects, “Effective Communication” or “Mabisang Komunikasyon,” merges all language and communication-related courses, including MIL.

Nazareno contrasted this with Finland’s integration of MIL in instruction from early childhood. The country consistently ranks at the top of the European Media Literacy Index.

“[There should be a] push from government and educational systems to do this even for people who are outside the school system like out-of-school youth, parents and the elderly since [they aren’t] digital natives,” he said.

He added that crafting adaptive, future-proof AI policies will require collaboration between the government, tech companies, academia and civil society.

Several AI-related bills aimed at upholding cybersecurity, regulating electoral disinformation and establishing an “AI Bill of Rights” were filed in the 19th Congress. Since these bills were not passed, legislators will have to refile them in the 20th Congress.

“I think AI will pose a bigger problem in the future. Steps should be made, not just by fact-checkers, but also those who profit from AI, to ensure ‘yung contributions nila to technology ay nananatiling beneficial sa lahat ng aspeto… lalo na sa mundo na mayroong demokrasya hanging by a thread,” Balbedina said.

(I think AI will pose a bigger problem in the future. Steps should be made, not just by fact-checkers, but also those who profit from AI, to ensure that their contributions to technology remain beneficial in all aspects… especially in a world where democracy is hanging by a thread.)

--

Victoria Gail Vasco is an intern at Philstar.com.