TIAA’s AI flagged a $3M retirement withdrawal scam, then humans stopped the transfer

Artificial intelligence is making scams faster, cheaper, and more convincing. In just seconds, it can clone voices, generate persuasive emails, and produce realistic deepfakes that push victims to hand over life savings. But TIAA CEO Thasunda Brown Duckett argues the same tools can also help catch fraud early enough to prevent a financial wipeout.

Duckett’s example is specific. A 76-year-old TIAA customer attempted to withdraw his entire $3 million retirement portfolio, and TIAA’s AI tool flagged the request as suspicious and out of pattern. The AI did the first crucial move: it recognized that something about the withdrawal did not fit the customer’s normal behavior. Then a portfolio manager escalated it to TIAA’s fraud team, which spent hours trying to convince the customer he was being scammed, a hard human task because the first instinct after an apparent opportunity or instruction is not disbelief but momentum.

Why was the human part so decisive? Duckett described the psychological trap in plain terms: “The first thing is you don't want to believe you've been scammed. You've almost been trained to defend the scam.” That is the moment AI detection still needs a person. AI can flag, score, and route alerts. It cannot replace the difficult conversations where a fraud team has to slow the person down, challenge the narrative being pushed, and sometimes contact family members so the victim hears an outside perspective before the transfer completes.

In this case, the escalation worked. A fraud specialist contacted the man’s daughter, and TIAA stopped the money from moving. Duckett said the participant told the company, “You saved my bacon,” framing the outcome as both emotional and practical. This is also the point Duckett emphasized in conversation: AI is not just another tool in the scam arms race, it is part of the defense. She called the scenario an intersection of “your workplace, which is our people, your culture, and AI,” adding that AI “by itself would not have necessarily protected this person.” Her broader line was direct: “The human is not just in the loop, it's still all about the humans.”

The business stakes behind that claim are rising fast. Duckett tied the case to the scale of AI-driven fraud and broader cybercrime losses. She noted that AI is fueling over 1 million scams a year, reinforcing the need for humans and technology to work together. She also cited FBI data: over $20.8 billion was lost to about 1 million cybercrime complaints in 2025 alone, with older adults accounting for the largest number of complaints and total losses. For 2024, individuals over 60 lost $4.8 billion to scams, while those aged 50-59 lost $2.5 billion. And the comparison Duckett included underscores the acceleration: overall cybercrime losses totaled $1 billion in 2015.

For executives and boards, the second-order implication is uncomfortable but actionable: fraud prevention is becoming both more technical and more labor-intensive, even if AI automates detection. As scams get better at impersonation and persuasion, financial institutions will need systems that can identify anomalies in real time, plus teams trained to handle the human response time, emotional resistance, and escalation logistics that automated systems cannot reliably manage. That means operating models will shift from “monitoring alerts” to “orchestrating outcomes.” Detection is step one. The measurable success is whether money stops moving.

Duckett also connected this to the future of work, using the same logic to argue against simplistic automation narratives. At a time when AI-driven restructuring plans have spooked workers and some business voices have predicted entire job categories could vanish, Duckett said she does not believe work will disappear. She suggested that jobs are not going away, even if some roles do. In her view, areas tied to cybersecurity, fraud prevention, and AI oversight are likely to need more people, not fewer. She said, “Cyber is a big risk,” and that “We will need more people in cyber, we will need more people monitoring, we will need more people analyzing what AI is telling you.”

She also pointed to workforce dynamics where entry-level roles can evolve rather than disappear. Digital natives may have an advantage because they tend to be comfortable with emerging technologies and can help leaders interpret what AI can do and what it can’t. “Our young people have always informed leaders on where the puck is going,” she said, warning against ignoring the next generation. Finally, she urged leaders to handle the anxiety AI creates by having open conversations about what it means for individuals: “[AI] comes with a lot of anxiety,” and companies need “honest conversations.”

Put it together and the TIAA episode becomes a template for what financial services leadership will increasingly be judged on: can you deploy AI to detect anomalies like a 76-year-old trying to cash out a whole retirement portfolio for the first time, and can you back it up with real human escalation that changes the outcome before the transaction finalizes? That is where the competitive moat will form. Not in the model alone, but in the combined system of policy, process, people, and tech that turns suspicion into prevention.