Law Enforcers Focus on AI Threats, While Banks See a Fraud-Fighting Tool
AI-generated media can make impersonation scams more convincing, while banks are exploring whether AI can help detect fraud. Recent moves by U.S. officials and Interpol highlight the risks. A survey of Asia-Pacific banking leaders points to potential defenses, and the safeguards they still need.
- DHS ordered DeepSeek products removed from federal civilian agency systems.
- Interpol is using AI-generated videos to warn about deepfake scams.
- 60% of surveyed APAC banking executives selected fraud detection as generative AI’s strongest near-term opportunity.
- Privacy, regulation and explainability remain key adoption concerns.
These developments touch on different parts of the AI debate. The DHS order concerns national security and data handling. It does not allege that DeepSeek products were used in fraud. Interpol is warning about synthetic media used in scams, while banks are considering AI to detect suspicious activity. Each case shows why AI claims should be judged by the specific risks and uses involved, not by the technology’s promise alone.
DHS orders DeepSeek products removed from federal systems
On October 19, 2026, Homeland Security General Counsel James Percival II filed an order directing federal civilian agencies to remove DeepSeek products from their IT systems and barring affected agencies from making new purchases. The order covers DeepSeek affiliates and third-party products and services developed using DeepSeek implementations.
The stated concerns include data privacy, possible access by Chinese authorities and the risk that sensitive information could be transferred to China. DeepSeek is a Hangzhou-based AI company founded in 2023 and backed by Chinese hedge fund High-Flyer.
Its R1 reasoning model drew international attention in January 2025, after reports that it delivered competitive performance at a fraction of the development cost of leading U.S. AI systems. Its emergence was followed by a sharp technology-stock sell-off that wiped billions of dollars from Nvidia’s market value. President Donald Trump called the breakthrough a “wake-up call” for the U.S. technology industry.
U.S. scrutiny continued. In June 2025, Reuters reported that the government had been assessing DeepSeek’s links to Beijing since the company’s inception. The White House unveiled its AI Action Plan the following month. In September 2025, the Center for AI Standards and Innovation (CAISI), part of the National Institute of Standards and Technology, published an evaluation that raised concerns about DeepSeek’s security vulnerabilities and political bias. DeepSeek’s chip-ban controversy has also added to U.S.-China technology tensions.
The order restricts federal systems. It does not prove that every DeepSeek product or user has been compromised. The available details do not explain how the order will be enforced or identify every affected agency and product. National-security precautions can be warranted, but their basis and scope should be clear, especially when they affect a foreign competitor in a strategically important industry.
Interpol warns that deepfakes can undermine familiar scam checks
On October 19, 2026, Interpol announced a public-awareness campaign featuring three AI-generated videos that show how synthetic media can be used in scams. Deepfakes, or AI-generated or altered media that imitate real people, can feature in investment fraud, impersonation scams, romance scams, sextortion, deepfake pornography and disinformation.
“Generative AI applications and the large language models that power them are becoming more sophisticated by the day, no longer requiring significant technical skills or investment, ” Interpol said. The organization also warned that tools can reproduce a person’s face or voice using a small amount of material found online. A fake’s credibility can depend on the quality and quantity of that material, the tools used and how the impersonation is presented.
As a result, some familiar warning signs are becoming less reliable. “AI is erasing the tell-tale signs of a scam. Spelling mistakes, branding inconsistencies, clumsy formatting: these red flags are fast becoming a thing of the past, ” said Toshinobu Yasuhira, director of the Interpol Innovation Centre. A polished message no longer proves that its sender is legitimate.
Interpol’s advice is simple: “Wait. Validate. Don’t take the bait.” Verify unexpected calls and urgent requests through an independent channel instead of relying on the contact details or links provided. Agreeing on a “safe word” with friends and family can help confirm whether an unusual call is genuine. Interpol also advises keeping software updated and watching for unsolicited contact, emotional pressure, unusual payment methods and virtual relationships that move unusually fast.
Entrust’s 2026 report examined more than 1 billion identity-verification events across 195 countries. It found that deepfakes accounted for one in five biometric fraud attempts and that injection attacks increased 40% year over year. An injection attack tries to feed manipulated images or video directly into an identity-verification system instead of presenting them through a live camera capture.
Those figures do not mean that one in five fraud cases overall involved a deepfake. They refer to biometric fraud attempts in Entrust’s reporting. The available details do not specify the precise category definitions or whether the attempts were detected, confirmed or measured in another way.
Interpol’s broader response includes Project SynthWave PLUS, launched in March 2026 to strengthen law enforcement’s ability to counter synthetic-media threats, particularly in Southeast Asia. Public awareness is one part of that effort. Investigators also need the capacity to recognize and respond to manipulated media.
Banks see potential, but adoption hinges on trust
In a FICO poll conducted at its Asia Pacific Banking Leaders Forum in Bali in August 2026, 60% of surveyed banking executives said fraud and scam detection offered the greatest potential for operational improvement from generative AI over the next 12 months. The poll gathered responses from 51 senior executives and C-suite leaders from leading banks across APAC. TNGlobal published the results on October 16, 2026.
The result reflects the views of those respondents. It is not a comprehensive measure of every bank in the region, nor evidence that AI has already reduced fraud at scale. FICO reported percentages, but the available poll details do not specify the response base for each question. The figures should be read as reported shares, not precise counts of executives.
Data privacy and protection was the most frequently cited barrier to AI adoption, selected by 42% of respondents. Regulatory and compliance risk and model transparency and explainability each accounted for 21%, while 16% cited computing costs or infrastructure.
These concerns go beyond whether a model can flag a suspicious transaction. Banks may need to explain why it flagged one, show that the system can withstand an audit and protect sensitive customer data. “Financial institutions operate under regulatory scrutiny that most consumer technology companies don’t face, ” said Dattu Kompella, FICO’s managing director in APAC.
“Banks need AI models that are explainable, auditable and designed from the ground up for compliance. Off-the-shelf consumer AI doesn’t meet that bar.”
The survey also found that 77% of respondents considered it extremely important for banking AI tools to be built for financial services. The remaining 23% said specialization was somewhat important. While 26% reported that task-specific AI models were already deployed, 46% expected deployment within one to two years.
Purpose-built systems may be better suited to financial oversight, as Kompella argues, but the label alone does not guarantee safety or performance. A model designed for banking can still make errors, expose data or produce decisions that are hard to explain. Banks will need to test systems under real operating conditions and assess their governance, not simply trust a vendor’s pitch.
Key questions about AI fraud risks and defenses
-
Why did DHS restrict DeepSeek on federal civilian systems?
The order cites concerns about data privacy, possible access by Chinese authorities and the potential transfer of sensitive information to China. A September 2025 CAISI evaluation also raised concerns about security vulnerabilities and political bias. The available findings do not establish that every DeepSeek product has been compromised.
-
How can people respond to a possible deepfake scam?
Pause and verify through a separate, trusted channel. Watch for urgency, emotional pressure and unusual payment requests. You can also agree on a safe word with family or friends.
-
What does Entrust’s deepfake statistic measure?
Entrust reported that deepfakes accounted for one in five biometric fraud attempts in its 2026 report, which examined more than 1 billion identity-verification events. The figure does not represent deepfakes’ share of all fraud, and the available details do not specify the exact category definitions.
-
Where do APAC banking executives see AI’s greatest near-term value?
Fraud and scam detection. FICO reported that 60% of surveyed executives selected it as the area with the greatest potential for operational improvement over the next 12 months.
-
What concerns are holding banks back from adopting AI?
Data privacy and protection led at 42%. Respondents also cited regulatory and compliance risk (21%), model transparency and explainability (21%), and computing costs or infrastructure (16%).
AI needs safeguards, not blind faith
AI can help criminals create convincing impersonations. It can also help institutions analyze suspicious activity. Neither outcome is automatic. The technology’s value depends on how it is used, the quality of the data and the safeguards around the decisions it informs.
For individuals, that means independently checking high-stakes requests instead of trusting a familiar voice or polished message. For banks, it means demanding privacy protections, explainable decisions and proof that systems work under real conditions. AI may help catch scams, but it cannot replace sound security or accountability.