AI agents, which are autonomous software tools designed to perform tasks for users, are rapidly entering everyday life. These tools offer convenience, handling tasks from grocery ordering to fund transfers, by interacting with digital platforms in ways similar to humans.
However, the wider adoption of AI agents presents challenges, particularly in the realm of fraud detection, potentially undermining existing security frameworks.
The evolving digital landscape
More than 60% of visitors to online shopping platforms are now bots rather than humans
Recent trends highlight the growing presence of bots within the retail sector. More than 60% of visitors to online shopping platforms are now bots rather than humans.
At the same time, over half of consumers have expressed comfort with AI agents performing online shopping and browsing on their behalf, a figure that continues to rise. Furthermore, the traffic generated by generative AI on US retail and banking sites has increased by 2000% in just the past year.
The challenges for fraud detection
Traditional fraud prevention mechanisms operate under the assumption that interactions are human-driven, an assumption that AI agents continually challenge. AI agents interfere with several layers of detection:
- Device Fingerprinting: Since agents often use cloud servers, the link between user and device identity is weakened, making device signals less reliable.
- Behavioral Biometrics: Unlike humans, agents do not exhibit typical user behaviours, such as hesitations or mistyping, making it difficult to distinguish between legitimate and malicious activity.
- Bot Detection: Designed as bots, these agents make it hard for fraud teams to differentiate between a potential fraudster and a legitimate customer.
The consequence is that fraudsters can exploit legitimate AI agents to circumvent detection, effectively operating undetected.
The increasing implications
Fraud detection teams are currently facing workloads that have increased by 2–3 times
The potential financial impact is significant, with fraud losses expected to rise by as much as 500% as attackers utilise AI-driven automation.
Deloitte projects that losses in the US financial services sector could escalate to $40 billion by 2027, from $12.3 billion in 2023. Fraud detection teams are currently facing workloads that have increased by 2–3 times just to sustain existing levels of security.
Adapting to the new reality
Conventional detection methods, including device intelligence, behavioural analytics, and bot detection, are becoming less effective. To stay ahead, there is a need for new fraud prevention strategies that focus on predictive and agent-aware AI models. These models must understand and evaluate intent, not just the automation processes themselves, and adapt in real-time to different types of agent traffic.
The advancement in fraud detection involves consolidating risk assessments across multiple sessions and transactions without increasing customer inconvenience. By implementing agent-aware models, financial institutions and retailers can utilise AI to address security threats effectively, maintaining customer satisfaction without compromising on protection.
