Human Emotion of Fear; The Psychology of Phobias and The Little Albert Experiment

FACES OF DECEPTION
LEGAL AND ETHICAL CONCERNS OF DEEPFAKES

A research Paper by: Eshaal Ammar, Maheen Faheem, Bazil Jalal, Mahd, Sonia, Purvi

Abstract:

Generative deep learning algorithms have progressed to a point where it is difficult to tell the difference between what is real and what is fake. Deep fakes are a form of synthetic media generated through deep learning algorithms, that convincingly imitate real people, manipulating audio, images, or video to create highly realistic but fabricated content. Deepfakes present both innovative opportunities and major risks. This paper examines (1) what deep fakes are, (2) whether and how deep fakes infringe IP and publicity rights. And furthermore, it simultaneously (3) helps to understand the effects of deep fakes caused in society and challenges possessed due to misinformation, identity theft and fraud.

 

Introduction:

“CEO: Sir, please transfer me $10,000 from accounts, we are going to sign these documents." The director, who has spoken with the CEO every day for years, hears the same tone and phrasing, only to learn later the voice was synthetic and now the money is gone. That scenario is not theoretical: audio deep fakes have been used in real frauds and are now cheaper and easier to create than ever. A deep fake is an image or recording that has been convincingly altered and manipulated to misrepresent someone as doing or saying something that was not actually done or said. They are typically created using Generative Adversarial Networks (GANs), which pit two neural networks against each other to produce increasingly realistic fakes.

  Detection researchers classify tasks like face-swapping, talking-face generation, and voice cloning and measure both generation quality and detection difficulty.

This paper asks: To what extent do deep fakes generated by AI endanger digital privacy and online security? I treat the question across two axes:
(a) individual harms (privacy, defamation, non-consensual imagery, and impersonation/fraud)
(b) systemic harms (election misinformation, erosion of trust in media, enabling organized crime).

 

Literature review:

We reviewed 3 literature review/survey documents, published in 2022. The majority included deepfake generation and detection techniques and the remaining focused only on deepfake detection techniques. We reviewed 3 literature review/survey documents, published in 2022. The majority included deepfake generation and detection techniques and the remaining focused only on deepfake detection techniques.

Almutairi and Elgibreen (2022) identified three types of audio deepfakes: synthetic-based, imitation-based and replay-based. This survey examined machine and deep learning audio deepfake detection technologies and determined that machine learning methods proved more accurate than deep learning, but required excessive training and manual feature extraction.

Malik et al. (2022) took a similar approach, evaluating face image and video deepfake techniques (generation and detection) up to early 2021, and concluded there was a general inability in the detection models to transfer and generalise indicating further research was needed.

Masood et al. (2022) surveyed publications relating to audio/video manipulation, specifically generation and detection techniques available up to March 2022. It found some of the limitations of generation processes were generalisation across source datasets and pose variations. While the challenges identified with deepfake detection technologies were the quality, fairness, and trust of deepfake datasets, temporal aggregation, and social media laundering. Generation and detection techniques available up to March 2022. It found some of the limitations of generation processes were generalisation across source datasets and pose variations. While the challenges identified with deepfake detection technologies were the quality, fairness, and trust of deepfake datasets, temporal aggregation, and social media laundering.

Risks:

Privacy Risks

Deepfakes can pose a serious threat to confidentiality by enabling bad actors to impersonate individuals in order to gain unauthorized access to sensitive information .Similarly, explicit videos of an individual can be made, which may then be exploited for purposes such as blackmail, coercion, or extorting money in exchange for their removal.

Security risks

Deepfake voice phishing (also known as Deepfake vishing) is a problem in which a family member will receive a call that sounds like their loved one urgently requesting money due to an emergency. Or a person with access to a company’s finances may receive a call that sounds like their CEO instructing them to transfer money to an account for a “business deal.” However, instead of their family members or their CEO, they are transferring the money to a scammer. e.g. group-IB.

Deepfakes can be used in various ways to create convincing fake news articles, videos or audio clips of candidates doing controversial or damaging things. They are then in turn used to manipulate public opinion, sway voter preferences or even suppress turnout. E.g. cloud secure alliance .Similarly, Deepfakes can be used to create fake identities or impersonate real individuals, facilitating various forms of identity theft and fraud. For example, deepfakes could be used to create fake social media profiles, generate false evidence to support scams or phishing attempts, or even bypass some biometric authentication systems. (cloud security alliance)

 

 Right of publicity and personality rights

Unconsented use of likeness/voice is often actionable under right-of-publicity regimes, which protect commercial exploitation of a person’s persona; courts and plaintiffs have used these claims against AI-based impersonations where commercial exploitation or reputational harm is present. The George Carlin estate litigation and settlement over an AI-created comedy special is a concrete example. Reuters A.P News

 

Social & operational mitigations

Stronger authentication & operational controls. Organizations must assume media can be faked and adopt multi-factor and out-of-band verification (e.g., codewords, direct confirmation channels) to blunt social engineering. Financial institutions, HR teams, and executives need updated playbooks and staff training. MDPI Business Insider

Media literacy and rapid verification teams. Rapid response, public literacy, and forensic hubs (law enforcement + newsroom labs) reduce the spread and impact of viral deepfakes. EuropolTIME

Countermeasures and Solutions

Platforms like Microsoft’s Video Authenticator, and Deepware Scanner Mobile can spot deepfakes by noticing problems like facial inconsistencies, unusual blinking, unusual emotions, and artifacts or inconsistencies in the audio, which can help people identify if the content they see is real or fake. (privacy end).Digital watermarking, such as hashing, can provide a video file with a short string of numbers that is lost if the video is manipulated. It can also provide an authenticated alibi for public figures, given that they constantly record where they are and what they are doing (timreview.ca).

Another solution is to conduct public awareness campaigns and structured training initiatives to provide people with the skills to recognize a deepfake. These programs will help them understand clear signs like unnatural facial movements, mismatched lip sync, and oddly robotic or unnatural audio.

Conclusion

Summarize key insights: deepfakes are a significant and evolving threat to digital privacy and security, and although IP rights offer some defences, they’re not enough on their own. Call for stronger legislation, technological innovation, and media literacy to protect individuals and institutions in the AI age.