AI clones: the good, the bad, and the ugly

The ability of artificial intelligence to mimic human voice, appearance, and personality has moved beyond simple novelty into a complex ethical landscape. While the industry has established a baseline for consensual AI cloning, a new wave of applications is creating digitally replicated personas without the consent of the subjects, blurring the lines between productivity tools, emotional support, and harassment.

At the most transparent end of the spectrum are digital twins created by public figures for communication and accessibility. Executives and politicians have begun deploying AI clones—combining chatbots with visual avatars—to interact with audiences on their behalf. Mark Zuckerberg and LinkedIn co-founder Reid Hoffman have both explored or implemented digital versions of themselves to scale their personal interactions.

Political figures have used similar technology for strategic outreach. Former Pakistani Prime Minister Imran Khan utilized an authorized voice clone to campaign while imprisoned, and New York City Mayor Eric Adams employed voice-cloned robocalls to communicate with constituents in languages including Mandarin and Yiddish. These use cases are generally viewed as ethical provided the recipients are informed they are interacting with an AI.

However, the technology has been weaponized for high-stakes fraud and extortion. Non-consensual cloning has enabled scammers to bypass traditional security instincts by mimicking trusted voices and faces.

In 2019, the first widely documented instance of this occurred when scammers mimicked the voice and German accent of a parent company executive, tricking the CEO of a UK energy firm into transferring €220,000 to a fraudulent account. In 2023, an Arizona mother, Jennifer DeStefano, was targeted by extortionists who used an AI clone of her 15-year-old daughter’s voice to demand a $1 million ransom.

The sophistication of these attacks increased in 2024, when a finance worker at a multinational firm in Hong Kong transferred $25 million after attending a video conference call. The call featured deepfake recreations of the company’s CFO and several other colleagues, demonstrating that real-time visual and auditory cloning can now deceive professional financial controls.

The Rise of Workplace Persona Cloning

A more nuanced ethical challenge has emerged through the creation of “colleague clones.” This trend, prominent in China, involves employees using specialized software to build digital versions of their bosses or coworkers to navigate office dynamics or preserve information.

A central driver of this trend is a project called Colleague Skill, released in March 2024 by Shanghai-based engineer Zhou Tianyi. The tool and its open-source derivatives allow users to upload chat histories, emails, and internal documents to create a functional persona that mimics a specific individual’s professional expertise and communication style.

The technical architecture of these clones relies on a sophisticated stack of existing AI tools, including:

• Large Language Models such as Claude, ChatGPT, and DeepSeek API.
• Specialized models like Kimi.
• Optical Character Recognition (OCR) via Tesseract for processing documents.
• Sentiment analysis modules to capture emotional tone.

Unlike general-purpose AI, which acts as a broad knowledge base, these tools function as a specialized mask, forcing the AI to behave like a specific person. While Colleague Skill began as a satirical commentary on AI-driven layoffs, some employees have used it to retain institutional knowledge or as a sounding board to predict how a manager might react to a specific proposal. In most reported cases, these clones are created without the subject’s consent.

Digital Resurrection and Personal Cloning

The application of persona cloning has extended into the personal sphere with the development of Ex-Partner Skill, a fork of the Colleague Skill engine. This tool allows users to re-create former partners by uploading photos, social media posts, and chat logs.

The resulting chatbot mimics the former partner’s tone, catchphrases, and shared memories, allowing users to simulate conversations with someone no longer in their life. This technology has also been adapted into “deathbots,” where users create simulations of deceased loved ones to find closure or say things they were unable to say in life.

The ethics of these tools exist in a grey area. Proponents argue that because the interactions are private, they do not constitute harassment or stalking, but instead serve as a therapeutic tool for emotional healing and reflection.

Critics argue that this represents a fundamental lack of consent. The use of personal data created within a relationship for a purpose the other person would find objectionable draws parallels to the non-consensual nature of revenge porn, though the medium is conversational rather than visual.

This trend is not limited to China. Users of the platform Character.AI have attempted to create bots based on ex-partners, leading the company to update its Terms of Service to explicitly ban the creation of bots using the likenesses of private individuals without their permission.

As AI continues to refine its ability to replicate human nuance, the industry faces a growing challenge in defining the boundaries of digital identity and the right to one’s own persona after a relationship—or a life—has ended.