Openai CEO Sam Altman has expressed concern about how emotionally dependent young people are being made on chat. Speaking at a Federal Reserve Conference this week, Altman said that many people are turning to AI Chatbott for all types of life decisions, in the ways that go much further than accidental use. “People rely too much on chat,” Altman said. “There are young people who say such things, ‘I cannot take any decision in my life, without telling everything. I feel very bad.”

He called it “actually common” among young users, warning that such over-dependence could be harmful.

Altman’s comment is amid wide public debate how much people should rely in AI. Jeffri Hinton, widely known as AI’s Godfather, recently admitted in a CBS interview that he regularly uses the GPT -4 model of OpenIA and “goes to believe what it says,” even when he knows better. Hintan has spent years warning about the dangers of Superintendent AI, including manipulation and the risk of misinformation. Nevertheless, he confessed, “I am still trusting myself more on its answers.”

While AI is helping millions of people with tasks such as writing, abbreviation and coding, they are not infallible. In his test of Hinton, GPT-4 failed to give a correct answer to a basic argument puzzle. Despite this, he is ready for his convenience and speed.

But Altman’s warning did not stop at emotional dependence. He also raised an alarm about how AI is being misused by cyber criminal, especially in the financial sector. “AI is helping humans in incredible ways,” he said. “But all this help is not good.” During the same conference, Altman urged banks and financial institutions to be “smarter than the smartest AI” to protect the customer’s money from increasing threats such as voice cloning and video deepfack.

“One thing that hurts me is that there are still some financial institutions that accept the voiceprint as authentication,” he said. “This is still a crazy thing. AI has completely defeated in the ways that people currently certify.”

Ultman said that the AI-related voice clones are now almost sufficiently enough to copy people. This voice-based certification can make it useless. He warned that the video deepfac would be the next major threat, which is likely to target facial identification systems.

“I am very nervous that we have an important, adjacent fraud crisis,” he said. “Right now, this is a voice call. Soon, it is going to be a facetime that is uninterrupted from reality.”