Chapter 101: Autonomous Agents and Personalized Scams — The Digital Hunter That Never Rests

Throughout the years, the world of digital fraud resembled a wide fishing net cast into the sea in the hope of catching something. 'Nigerian Prince' emails, SMS messages about a package delayed in the mail, or fake bank links—all were generic, often written in poor Hebrew, and sent to many people simultaneously using a 'shotgun' approach. Those who kept their eyes open, those who exercised basic caution, managed to evade the net with relative ease. But as one examines voice and video fakes, a much quieter and more dangerous revolution can be seen: the transition from 'net fishing' to 'laser-guided harpoon hunting.'

Meet the autonomous agents. These are no longer simple bots that respond to keywords, but artificial intelligence-based software capable of setting goals for themselves, planning strategies, gathering information from every corner of the web, and maintaining a relationship that lasts for weeks and months—all without human contact on the other side. If in the past a fraudster had to invest time and energy to deceive a specific victim (known in professional jargon as Spear Phishing), today, the autonomous agent can do this for many victims simultaneously, with a level of precision and manipulation that was once reserved only for skilled intelligence agents.

In this chapter, we will analyze this technology, understand how it may affect psychological defense mechanisms, and examine the legal tools that may be available to victims when the adversary is code designed to inflict financial damage.

The following scenario is an educational example based on common fraud patterns. The characters and events are for illustration purposes only.

The Case of Yoram: When an Algorithm Becomes a Confidant

Yoram (pseudonym), 54, VP of Operations at a large logistics company, is considered a level-headed person. He does not click on suspicious links, he uses two-factor authentication on all his accounts, and he adheres to his bank's security rules. One morning, he received an inquiry on LinkedIn from a woman named 'Naama Levy,' who introduced herself as a market analyst in the supply chain field. Her profile looked reliable: a professional photo (created by AI, but it was difficult to tell), recommendations from colleagues in the industry, and a list of publications.

'Hello Yoram,' Naama wrote. 'I read the interview you gave to Calcalist last month regarding the raw materials shortage. Your analysis regarding the Eastern ports was fascinating. I am working on a similar study and would love to hear your opinion on a specific point.'

Yoram was flattered. The conversation developed slowly. For three weeks, 'Naama' asked for nothing. She sent Yoram relevant articles, asked professional questions, and demonstrated deep knowledge. She knew when he was going on vacation (information published online), she knew the names of his children (from an old social media post), and she always wrote in fluent, substantive Hebrew. What Yoram could not have known is that 'Naama' is a possible scenario of an autonomous agent—a sophisticated script connected to a Large Language Model (LLM) that scans the digital footprint in real-time to build an emotional manipulation.

The crisis arrived on Tuesday evening. Naama sent an urgent WhatsApp message: 'Yoram, I'm in terrible trouble. My startup won a government tender, but due to a technical glitch in an international transfer, our guarantees didn't go through and the tender will be canceled tomorrow morning. I am short 150,000 shekels for just 48 hours. I don't dare ask, but you are the only person in the field I trust. I have all the documents confirming the win.'

She sent PDF files that looked official. Yoram, who felt he had known Naama for a month and viewed her as a colleague, did not suspect. He transferred the money via a wire transfer to an account she provided. The next day, the LinkedIn profile of Naama disappeared, the WhatsApp number was disconnected, and the money moved through a rapid chain of accounts until it became cryptocurrency.

In such a scenario, when a victim turns to legal counsel, he may express deep frustration and wonder how a machine managed to simulate a human personality so convincingly, using accurate personal and professional information.

The Evolution of the Trap: From the 'Net' Method to the Autonomous 'Harpoon' Method

The Transition from Static to Dynamic

The fundamental difference between the frauds of the past and the frauds of autonomous agents lies in dynamism. In the past, a fraudster would prepare a fake 'landing page' and wait. Today, the autonomous agent is active and reacts to the victim's responses. If the victim expresses doubt, the agent may change tactics. If the victim asks a complex question, the agent is capable of locating information and formulating a convincing explanation within seconds.

AI-Based Open-Source Intelligence (OSINT)

Autonomous agents often operate using 'Open-Source Intelligence.' They scan social networks, blogs, forum comments, and even business registries. The goal is to build a psychological profile: what are the target's weaknesses? What are their aspirations? The agent customizes the 'lure' personally, making the fraud difficult to identify.

Electronic Patience

Unlike a human actor who might show impatience, an autonomous agent can wait a long time. It can manage 'small talk' for months, mention personal dates, and send content recommendations, only to build the trust required for the moment the financial request is made. This patience is intended to neutralize natural initial suspicion.

The Machine Behind the Mask: How Autonomous Agents Change the Rules of the Game

Technologically, autonomous agents operate in feedback loops. The attacker may define a general task of achieving financial gain from a certain target audience, and the agent acts independently:

1. Task Decomposition: The agent identifies a list of targets, builds a reliable identity, and creates initial contact.
2. Execution: Using digital tools to locate targets, creating a backstory via AI, and sending personalized inquiries.
3. Learning and Adaptation: Analyzing responses and improving the approach according to the interaction with the target.

The ability to carry out a 'Mass Personalization' impersonation attack is one of the most significant threats today. A single server can run a large number of such agents simultaneously, with each one managing a unique communication with a different target.

'Pig Butchering' Scams in the Autonomous Era

This type of fraud is based on gradual trust building and expectation of profits. Autonomous agents may guide victims to open accounts on fictitious investment platforms, display non-existent profits, and even allow the withdrawal of small amounts initially to strengthen trust—all in an automated process.

The Legal Struggle in the Digital Fog: Responsibility of Banks and Financial Institutions in the AI Era

A central question that arises in these cases is the issue of responsibility for the loss of funds. In a world of autonomous agents, the answer requires an examination of the existing law in Israel.

Payment Services Law, 5779-2019

This law determines the division of responsibility between the customer and the financial institution in the case of 'unauthorized use' of a payment method. As a rule, for a transaction not authorized by the customer, the responsibility lies with the financial institution. However, when the customer executes the transfer themselves due to manipulation, the legal situation becomes more complex.

In such situations, the issue of the bank's heightened duty of care arises. One can examine whether the bank has an obligation to identify anomalous patterns. If the systems did not identify a significant transfer to a party with whom there was no prior contact as a suspicious event, it may be possible to examine the liability of the financial institution under the circumstances.

The Issue of 'Gross Negligence'

The law may exempt a financial institution from liability if the customer acted with 'gross negligence.' The legal question is whether falling victim to a sophisticated autonomous agent, which simulated a human persona over time, constitutes such negligence. There is a professional position arguing that as technology advances, the threshold of expectations of the reasonable person may change, and the level of protection required by financial institutions must be examined against advanced, technology-based manipulations.

The Responsibility of Digital Platforms

Another legal issue concerns the liability of social networks that allow the activity of autonomous agents. Despite broad protections existing for platforms regarding user-generated content, legal possibilities can be examined in cases where the platform did not act after a warning or when an recommendation algorithm contributed to the creation of the initial contact.

Warning Signs: How to Identify an Autonomous Agent Before It Attacks

Despite the sophistication, one can sometimes identify cracks in the digital identity of autonomous agents:

• Response Speed: Very fast replies to long and complex messages at any hour, in a way that is not always natural.
• Emotional and Communication Nuances: AI agents may have difficulty identifying cynical humor, deep sarcasm, or very specific cultural cues.
• Avoiding Meetings: Repeated excuses regarding the inability to hold a live video call or a face-to-face meeting.
• Changing the Nature of the Relationship: A relationship built over time without physical contact, which suddenly shifts to requests of a financial nature.
• Profile Characteristics: Images with an artificial look, lack of natural social interactions, or a list of connections that appears random.

Possible Steps in Case of Suspicion of Fraud

1. Ceasing Communication: Immediate disconnection of contact with the suspicious party to prevent further manipulation.
2. Documentation: Saving screenshots of the conversations, profiles, and documents, as this information may be deleted by the other side.
3. Contacting the Bank: Urgent contact with the bank's security department to examine the possibility of stopping transactions, insofar as this is possible under the circumstances.
4. Reporting to Authorities: Filing a complaint and mentioning the suspicion of the use of artificial intelligence technology, which assists in characterizing the event.
5. Legal Examination: Examining the financial institution's compliance with its duties and whether there are grounds to examine the recovery of funds.

Legal Aspect: Liability and Restitution

In cases of personalized fraud, the legal path is examined according to the specific circumstances of each case.

Professional Liability Aspect: If fraud was carried out using information that leaked from an employer or a service provider, liability can be examined according to the Privacy Protection Law.

Duty of Disclosure and Warning: There is a legal approach according to which courts may examine whether the bank provided sufficient warnings in real-time. A general legal analysis suggests that under certain circumstances, if a financial institution is aware of common fraud patterns and does not warn a customer performing a similar transaction, this may be examined through the lens of tort law and negligence.

In conclusion, in an era of machine learning technology, the best protection combines awareness and caution together with the assertion of legal rights. During years of professional work, the need arises to examine how the legal system adapts itself to this changing reality, to ensure optimal protection for the public.