AI face swapping video could be a bonanza for scammers

Welcome to AI Decoded, Fast Company’s weekly newsletter that breaks down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company,covering emerging tech, AI, and tech policy.

This week, I’m focusing on a new class of AI video generation tools that could let scammers speak with a completely different face and voice on video calls. I also look at how Apple employees see Tim Cook’s appeasement of President Donald The President, and why OpenAI models got worse at writing. 

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X @thesullivan. 

A scammer’s dream

Video generation models have been improving over the past several years. We’ve always known that things could get weird when generative video becomes nearly indistinguishable from real video, and we’re entering that phase now. We’ve already seen convincing political and sexual imagery deepfakes. Another devious application of AI generative video will be phishing scams.  

Over the past six months or so, people have been posting on X about new AI tools that are capable of “face swapping” in real-time video. For example, the AI might make it look like it’s Leonardo DiCaprio or Scarlett Johansson saying my words and performing my facial expressions during a Zoom call. The AI analyzes the user’s facial movements and vocalizations in real time and sends them out via the faces of another person entirely. 

Some of these AI tools can generate a new face overlay using a single still image. So it may be possible for a scammer located anywhere in the world to find an image of a person on social media, use the AI to develop a reasonably convincing AI “face” from it, and then call one of the person’s relatives asking for money. The scammer might use a sample of the person’s voice to simulate the sound of the AI overlay character. 

Of course, the first people to be targets of such deepfake calls would likely be the elderly. They might be happy to hear from a grandchild, for instance, and perhaps less likely to question the authenticity of the video. They may not even know that such a simulation is possible. There’s a special place in hell for the perpetrator of such a scam, but it appears to be technically feasible.

How Apple employees see Tim Cook’s relationship with The President

Like other tech leaders, Apple CEO Tim Cook has chosen to engage with, rather than ignore, Donald The President. Cook has never publicly endorsed The President or donated to his presidential campaigns. But he did personally donate to The President’s inauguration fund, has visited the White House numerous times, and is in at least semi-regular contact with the president. 

During The President’s first term, Cook spoke out against the Muslim travel ban, the transgender military ban, and the president’s “good people on both sides” response to a deadly white-supremacist rally in Charlottesville, Virginia. In contrast, Cook has been quiet about the more extreme policies and actions The President has brought to his second term. Until now. This week, Cook was forced to speak out, at least to Apple employees. 

Cook came under fire from employees and customers for attending a private VIP screening of Amazon’s new Melania The President documentary at the White House on January 24, the same day that federal Customs and Border Patrol agents shot and killed VA hospital nurse Alex Pretti in Minneapolis. This followed a January 7 incident in which an Immigration and Customs Enforcement agent shot and killed Minneapolis resident Renee Good. 

Cook was silent about the Pretti shooting until the evening of January 27, when he reportedly sent a memo to Apple employees saying he is “heartbroken” about the Good and Pretti deaths and calling for “de-escalation.” He also said he’d had a conversation on the matter with The President. It may have had an effect. On Wednesday The President tried to calm the backlash against ICE. “We’re going to de-escalate a little bit,” he said, using the same term Cook did. (No actual de-escalation has taken place.)

I suspect Cook’s relationship with the president is purely transactional and strictly business. As people, Cook and The President couldn’t be more different. 

Based on conversations I’ve had with people who work at Apple, the internal narrative is that Cook is dealing with The President in the way that’s most likely to ward off tariffs on iPhones, new regulations, and government probes of Apple’s various businesses. And, the narrative goes, Cook is speaking to The President in language The President understands. 

That explains the flattery and lavish gifts he’s brought to the White House, the most recent of which hits some of The President’s favorite notes. The commemorative glass piece (resembling a silicon wafer) is mounted on a 24-karat gold base and engraved with The President’s name. The president loves gifts, he loves gold things, and he loves his own name. 

As I wrote for Fast Company earlier this week, the shootings of U.S. citizens Good and Pretti in Minneapolis may put Big Tech leaders’ political alliance with The President to the test like never before. Even transactional friendships should be subject to some standards of trust and decency. If more violence happens, Cook’s calls with The President could become far less amicable. 

OpenAI sacrificed writing skill for coding skill in GPT-5

There was some backlash among users against OpenAI’s latest model, GPT-5.2. People felt that the new model had less personality, less of a voice of its own. And these shortcomings are reflected in its writing, which many users say is harder to read and less natural-sounding than that of earlier models. 

OpenAI CEO Sam Altman acknowledged this week that GPT-5.2 isn’t as good a writer as its predecessor GPT-4.5 was. “I think we just screwed that up,” Altman said during an internal “developer town hall” on January 26.  

With so much money on the line, I doubt it was a simple “screwup.” OpenAI was entertaining a different set of priorities when developing GPT-5.2. “We put most of our effort in 5.2 into making it super-good at intelligence, reasoning, coding, engineering, that kind of thing,” Altman said. “And we have limited bandwidth here and sometimes we focus on one thing and neglect another.” 

OpenAI may also have been responding to an industry-wide shift toward chatbots that are brainier but somewhat duller—that sound less like unconditionally supportive friends and more like responsible adults. There are safety and liability reasons for this. OpenAI is now facing multiple legal claims that its chatbot failed to push back when vulnerable users (including minors) began talking about plans to end their own lives. 

OpenAI may also have felt pressure from Anthropic’s increasingly popular Claude Code tool, used by many developers to generate production-ready software code. GPT-5.2, which now underpins OpenAI’s Codex coding tool, is better than earlier models at reasoning and coding skill.

What’s unclear is why OpenAI can’t have a model that’s both a strong writer and a good coder. Anthropic seems to have achieved that with Claude. 

More AI coverage from Fast Company: 

• Chip firm ASML to slash 1,700 jobs and posts record profit for 2025 thanks to AI boom
• Got an AI skill? Now you can prove it on LinkedIn
• Why Yann LeCun left Meta, and what it means for AI’s next frontier
• How K-12 schools are left on their own to develop AI policies

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.

How much The President violence will tech industry stomach before backing away

View the full article