Here's a descriptive answer to your questions, drawing on the provided

sources and our conversation history:

1. Understanding Deepfakes and Their Impact

 Definition and Forms of Deepfakes: Deepfakes are defined as

synthetic media created through AI algorithms to seamlessly
superimpose one person’s likeness onto another or to
manipulate or replace a person's appearance, voice, or actions.
The term "deepfake" itself originates from "deep learning," a subset of
AI that utilizes neural networks to generate and re-create realistic-
looking content. They are a product of advancements in artificial
intelligence (AI) and machine learning (ML), particularly
leveraging Generative Adversarial Networks (GANs) to create
hyper-realistic manipulated images, videos, and audio. Deepfakes can
exist in various forms, including videos, photos, texts, and voice
recordings. Their common characteristic is their potentiation to
deceive, tricking people into believing they are human-generated. In
recent years, deepfakes have rapidly developed to a point where even
individuals with good vision cannot distinguish between fake and
real items. This is because they rely on deep neural networks to
achieve a higher level of realism in visual content, requiring precise
features for consistency between video frames, making it challenging
to differentiate real from fake information.

 Societal and Individual Harms of Deepfakes: Deepfakes are

recognized as a potent means of disinformation. Beyond general
misinformation, they pose profound societal and individual harms
because they can blur our perception of truth and exploit our
natural tendency to trust visual evidence.

o Individual Harms: Deepfakes pose a serious threat to an

individual's ability to make informed decisions and
perceive the world. They can modify perception in virtual
environments, inhibiting an individual's ability to compare
experiences using common axioms. They exploit cognitive
heuristics, such as the tendency to trust visual and auditory
information, and emotional triggers, deceiving individuals
without requiring active engagement or scrutiny, making them a
significant threat to autonomy and epistemic integrity.
 Misinformation, particularly from deepfakes, can linger in
memory despite corrective efforts, exacerbating the "continued
influence effect". Deepfakes can lead to reputational damage,
create fake evidence, enable identity theft and blackmail (as
seen in a romance scam case), and be used for financial fraud
(e.g., impersonating executives for unauthorized transfers). They
can also cause personal harm through sexually explicit non-
consensual images of classmates or falsely portraying
individuals. During wartime, they make it impossible for
individuals to develop clear and unbiased opinions on complex
matters like war progress, directly impacting safety and lives.

o Societal Harms: Deepfakes threaten national security,

political stability, and media credibility. They can be used
for political manipulation (e.g., discouraging voters), distort
reality, and potentially violate consumer protection laws. A
significant effect is the undermining of trust in the
authenticity of footage, leading civilians to distrust any video
from a conflict, eroding confidence in media, and fostering
widespread epistemic uncertainty. This erosion of trust can
extend to government institutions, providing fertile ground for
conspiracy beliefs and unhealthy skepticism.

 Deepfake Use in the Russo-Ukrainian War and Future

Implications: The Russo-Ukrainian war saw a notable increase in the
use of deepfakes, primarily by Russia, aligning with its established five-
pillar strategy of disinformation and propaganda.

o Examples of Use:

 A fabricated video of Ukrainian President Volodymyr

Zelensky surrendering to Russian forces, aiming to
undermine morale and persuade the international
community of Ukraine's impending defeat.

 A deepfake video depicting "The Ghost of Kyiv," a mythical

fighter pilot, using video game footage, blurring lines
between fact and fiction.

 A fabricated video of Vladimir Putin declaring peace

with Ukraine circulating online.
  Deepfakes used to discredit Ukraine's cooperation
with allies, such as Russian intelligence agents
impersonating Ukrainian Prime Minister Denys Shmyhal to
contact Haluk Bayraktar, CEO of Baykar Defense.

 Manipulated video calls impersonating Vitali Klitschko, the

mayor of Kyiv, to influence attitudes toward Ukrainian
refugees in EU cities.

 The Ukrainian government created an "educational

deepfake" depicting Vladimir Putin walking in Mariupol,
aiming to show war consequences and combat Russian
propaganda.

o Why they were not "highly successful": The deepfakes in

this conflict were often unsophisticated and easily
demystified. For example, the Zelensky deepfake was swiftly
debunked because the Ukrainian government had already
warned of such attempts. The Baykar Defense deepfake was
intercepted by Ukrainian intelligence and posted on Telegram.
One possible explanation for Russia's limited exploitation of
deepfakes' potential is a lack of technical skills or available
technology, or that their information propaganda, effective
domestically, doesn't work as well in democratic
countries. It might also be that Russia was testing capabilities
to understand potential and evaluate adversary's defense.

o Future Potential and Challenges: Despite their limited

success in this war, future deepfakes are expected to be
more effective, potentially undetectable, scalable, and
weaponized. The smaller and more subtle the alteration, the
more difficult it becomes for individuals to identify and for
regulatory bodies to sanction them. During war, when individuals
are focused on survival, fact-checking is impractical, making
deepfakes an even greater threat to lives and well-being.
Furthermore, deepfakes complicate the collection of reliable
evidence for war crimes, as claims like Russia's regarding the
Bucha massacre being falsified using AI demonstrate the
increased skepticism they cause.

2. National Regulatory Approaches

 Comparison of National Regulatory Frameworks: Several
countries have initiated national regulations to address deepfakes,
each with distinct focuses and mechanisms:

o United States of America (USA): The USA lacks a single

comprehensive federal deepfake regulation but has seen
significant progress with a mix of federal attempts and state-
level laws.

 Federal Efforts: The Malicious Deep Fake Prohibition

Act of 2018 aimed to penalize creators of fake videos
facilitating criminal or tortious conduct. The National
Defense Authorization Act (NDAA) of 2019 (Section
5709) mandated reporting on deepfake use by foreign
governments and its national security impact. The
DEEPFAKES Accountability Act of 2024 is being
processed, primarily focusing on preventing the use of
deepfakes to infringe upon individuals' sexual
integrity. The TAKE IT DOWN Act (April 2025) addresses
non-consensual intimate imagery, including AI-generated
deepfakes, providing removal mechanisms and perpetrator
accountability. The NO FAKES Act (April 2025) aims to
protect individuals' rights against unauthorized use of their
likeness or voice in deepfakes.

 State-Level Regulations: As of 2024, 14 states in the

USA have introduced regulations primarily focused on
deepfakes' influence on political elections. These often
involve requiring disclosure of AI use or implementing a
complete ban on AI-generated content within specific
timeframes before elections unless accompanied by
disclosure. For example, Virginia expanded its revenge
porn law to include deepfakes in 2019, and New York's
Senate bill S1042 (October 2023) criminalizes non-
consensual deepfake dissemination. California's
Defending Democracy Act (2024) targets deepfakes in
elections, requiring platforms to remove them within 120
days prior, with disclosure requirements extending beyond.

o China: China has implemented concrete regulatory measures,

notably the “Provisions on the Administration of Deep
Synthesis Internet Information Services (Draft for
 solicitation of comments),” released in 2022 and enforced in
January 2023. This broadly regulates synthetic content (an
umbrella term for text, images, audio, video, and virtual scenes
created with generative models), including deepfakes, AI-
generated texts, and advanced image enhancement. This
approach aligns with China’s policy of maintaining "correct
political orientation" in the media. It mandates enhanced data
management, transparency, content management, and technical
security from deep synthesis service providers.

o United Kingdom (UK): The UK implemented the Online

Safety Act (OSA) in October 2023. A specific section of the OSA
addresses the nonconsensual sharing of deepfake intimate
images, with a particular focus on children. It imposes legal
obligations on companies, especially social media providers, to
identify, mitigate, and manage risks associated with illegal
content, particularly content harmful to children.

 Limitations and Concerns with National Regulatory

Frameworks: Each national approach faces certain limitations and
concerns:

o USA Limitations: While state-level laws might prevent deepfake

spread during election periods, they would not cover
dissemination during wartime outside of election periods,
indicating a need for broader applicability.

o China Concerns: Although China's "Deep Synthesis Provisions"

effectively address deepfakes, they raise concerns regarding the
infringement of freedom of information and opinion,
potentially coming at the cost of fundamental rights and
freedoms due to their alignment with maintaining "correct
political orientation".

o UK Limitations: The effectiveness of the OSA may be uncertain,

as its primary focus on children and lack of broader
applicability limits its reach. The OSA also tends to focus on
advisory committees and media literacy rather than direct
regulation of deepfakes.

3. India's Context and Challenges

 India's Lack of Specific Legal Framework and Limitations of
Existing Laws: India currently lacks a specific legal framework or
direct point of legislation that comprehensively addresses deepfake
technology. Instead, it largely relies on existing laws, which offer only
partial remedies and fail to fully address the spectrum of harm
caused by deepfakes.

o Application of Existing Laws:

 IT Act, 2000: Sections like 66D (penalizing impersonation

through electronic communication), 66E (protecting
privacy by criminalizing unauthorized capturing of images),
and 67 (regulating obscene content) can be invoked. For
instance, 66D can apply to financial frauds and scams
involving AI-generated impersonation, 66E to deepfake
pornography, and 67 to explicit AI-generated content.

 Forgery Laws (IPC Sections 464, 465): These laws

criminalize the creation of false documentation with the
intention to deceive. They can sometimes apply if a
deepfake involves altered official documents,
contracts, IDs, or is used in financial/corporate fraud
(e.g., impersonating executives for financial approval) or to
create false evidence in legal matters. Deepfakes for
identity theft with deceptive gains can also fall under these
laws.

o Limitations of Existing Laws:

 Lack of Direct Recognition: Existing laws do not

explicitly define deepfake-related offenses, making
prosecution more difficult as they were not drafted with AI-
generated content in mind.

 Definitional Ambiguity: Distinguishing between

harmless AI-generated content (like movie special effects)
and malicious deepfakes (fake political speeches) is
challenging, as the intent matters. Crafting a legal
definition that considers intent and harm is complex.

 Insufficient Scope: They fail to address the broader

spectrum of deepfake implications, such as political
misinformation or widespread reputational damage.
 Deepfakes used for political propaganda or defamation
might not fit neatly under existing sections like 66D, 66E,
or 67. There's no clear provision for AI-driven
misinformation except broader laws like IPC Section 505
(public mischief).

 Lack of Tangible Forgery: Traditional forgery laws focus

on falsified physical or digital documents; deepfake videos
and audio files may not always fit neatly into these existing
legal brackets, especially if they do not involve falsifying
documents.

 No Clear Provisions for Consent/Harm Mitigation:

There are no clear legal mechanisms for victims to demand
immediate takedowns or compensation for damages
caused by deepfakes, nor specific penalties for AI
developers or platforms that fail to prevent misuse.

 Rapid Technological Advancements: Deepfake

technology evolves rapidly, and the legal framework
lags behind, making laws quickly outdated. The
increasing sophistication makes AI-generated content
harder to detect, complicating enforcement.

 Challenges in Defining and Regulating AI-Generated Deepfakes

in India: The challenges in defining and regulating deepfakes in India
are multi-faceted:

o Definitional Ambiguity: It's hard to distinguish between

harmless AI-generated content (e.g., movie special effects) and
malicious deepfakes (e.g., fake political speeches). The intent
behind the deepfake is crucial, but hard to legislate for
broadly.

o Rapid Technological Advancements: Deepfake technology is

evolving rapidly, making it difficult for lawmakers to create
specific regulations that can keep pace with real-time changes.
This makes detection and enforcement increasingly complex as
AI-generated content becomes more sophisticated.

o Balance of Free Speech and Regulation: In democratic

countries like India, overly broad laws risk infringing upon free
speech and creative expression. Some deepfakes, like
 parodies or historical recreations, might be protected under free
speech, making strict regulations cautious.

o Anonymity and Global Reach: Deepfakes can be generated

and spread across borders anonymously, making it difficult to
track and penalize offenders. Identifying the source requires
advanced forensic equipment, which is not always accessible to
law enforcement.

o Multiple Legal Domains: Deepfakes span various legal

domains, including privacy, defamation, cybercrime, intellectual
property infringement, and national security, making
comprehensive regulation a multifaceted challenge that existing
laws cannot fully cover.

 Landmark Cases and PILs in India: Several cases and Public

Interest Litigations (PILs) highlight the urgent need for a more specific
legal structure in India:

o PIL in Delhi High Court (Lok Sabha Elections): A PIL was

filed against the use of deepfake technologies in political
campaigns for Lok Sabha assembly elections, seeking directions
for social media intermediaries (Meta, Google, X Corp) to take
down and block deepfake content concerning political figures.

o PIL by Kanchan Nagar, Vikas Saboo, and Images Bazaar:

Another PIL in the Delhi High Court challenged the unauthorized
use of original artistic works by AI platforms, raising concerns
about misuse of copyrighted material and exploitation of Large
Language Models (LLMs) to create unauthorized deepfake
content targeting women. This petition demanded amendments
to the IT Act and Copyright Act to include AI-related offenses and
AI-generated content under laws against impersonation and
cheating.

o ***Rajat Sharma v. AICC & Ors. (2024)***: The Delhi High Court
granted an ad interim injunction against the removal of
defamatory tweets and doctored videos circulated by members
of the All India Congress Committee (AICC) against journalist
Rajat Sharma. The court noted that edited videos falsely
depicted Sharma using abusive language, emphasizing the need
to balance freedom of speech (Article 19) with the right to
dignity (Article 21).
 o Anil Kapoor vs. Simply Life India and Ors. (2023):
Bollywood actor Anil Kapoor filed a lawsuit for the unauthorized
use of his persona via AI-generated deepfakes. The Delhi High
Court granted an ex-parte ad-interim injunction, restraining
defendants from using his name, image, voice, or persona
without consent, highlighting the judiciary's proactive stance.

o Rajat Sharma vs. Tamara and Ors. (2024): This case

involved a petition by Rajat Sharma and advocate Chaitanya
Rohila concerning deepfakes. The court directed the central
government to consult with deepfake technology providers,
telecom service providers, victims, and intermediaries before
finalizing recommendations for detecting and removing
deepfakes, emphasizing a collaborative approach.

o ANI vs. Open AI (2024): Indian news agency ANI sued OpenAI,
accusing ChatGPT of using its published content without
permission for AI training. This case highlights copyright
infringement concerns related to AI-generated content and the
permanent storage of data in AI models.

4. Challenges of Enforcement and Liability

 Challenges in Attributing and Enforcing Laws Against Deepfake

Creators: Attributing and enforcing laws against deepfake creators is
particularly challenging due to several factors:

o Cross-Border Nature and Anonymity: Deepfakes can be

generated and spread across borders anonymously,
making it extremely difficult to track malicious actors and
penalize them. Many creators use VPNs, encrypted channels,
or fake identities.

o Sophistication of AI: Modern deepfake technology is capable of

producing highly realistic content, making it hard to
differentiate real from fake. Subtle manipulations or low-quality
deepfakes can also make detection very difficult.

o Rapid Dissemination: Deepfakes can go viral before they

are even detected, misleading millions before takedown
measures can be implemented.
 o Lack of Robust Detection Tools: Deepfake detection
technology often lags behind AI generation, and existing tools
struggle with real-time identification.

o Limited Legal Framework: Many countries still lack specific

laws against political deepfakes, which leads to inconsistent
and reactive enforcement rather than preventive measures.

o Challenges in Proving War Crimes: Collecting reliable

evidence for war crimes is already difficult, and deepfakes
further complicate this by potentially creating false narratives
or questioning the authenticity of evidence.

 Controversy Surrounding Liability of AI Developers and Social

Media Platforms: The role and liability of AI developers and social
media platforms in deepfake proliferation is highly controversial.

o Arguments for Holding Tech Companies Accountable:

 Tool Provision: AI companies develop and deploy the

tools used to create deepfakes; if these tools are misused
for harmful purposes like misinformation or fraud, they
should bear some responsibility.

 Content Moderation Responsibility: Social media

platforms serve as primary distribution channels. If they
fail to detect and prevent harmful deepfake content,
they could be seen as facilitating its spread, similar to their
accountability for hate speech or child exploitation. The
Digital Services Act (DSA) in the EU mandates social
media providers to remove fake news and mitigate risks
like disinformation.

 Negligence: If deepfakes cause serious harm (e.g.,

interfering with elections, financial fraud), and developers
fail to implement safeguards like watermarking or
detection algorithms, this could be seen as negligence
leading to lawsuits or regulatory action.

 Unlawful Processing of Personal Data: Some cases

have seen social media providers obliged to remove
deepfakes due to unlawful processing of the personal data
of the individual portrayed.
 o Arguments Against (or Challenges to) Holding Tech
Companies Accountable:

 Legitimate Uses: AI developers argue that their

innovations have legitimate uses in entertainment,
education, and accessibility, and a broad-spectrum ban
could hinder innovation.

 Freedom of Information/Artistic Expression: Strict

oversight could encroach upon fundamental rights like
freedom of information and artistic expression.

 Content Moderation Bias and Financial Interests:

Content moderation can influence users' perspectives,
serving to enhance visibility or censor content diverging
from platform policies and interests. Social media providers
are primarily motivated by financial interests, utilizing
targeting methods like "filter bubbles". Concerns arise
about whether the private sector should assume the role of
countries in safeguarding individuals' rights. Recent
announcements, like Meta's decision to end fact-checking
in the USA, limit the success of this solution.

 Limited Efficiency: Holding platforms accountable for

unlawful data processing is efficient only in specific cases
and cannot be treated as a long-term solution.

 Constraining AI Development: Imposing responsibilities

on AI model developers to prevent fake material
generation could constrain the creative abilities of AI
models and hinder commercial exploitation of research.
Past attempts to halt AI development due to risks have not
been long-term solutions, as financial profit often
outweighs the benefits of such restrictions.

5. Proposed Solutions and "Cognitive Liberty"

 Regulatory Approaches: Benefits and Limitations:

o 1. Criminalization of Deepfakes:

 Benefits: This approach advocates for both criminal and

civil liability for creators and disseminators. It enables
individuals to sue for damages and penalizes
 nonconsensual sharing of intimate deepfakes and
malicious creation without proper labeling.

 Concerns/Limitations: Criminalizing deepfakes raises

significant concerns about potential infringements on
freedom of information and artistic expression. The
example of Russian "fake news" laws shows how
criminalization can lead to an exodus of independent
media and distort public perception, treating journalists as
criminals.

o 2. Transparency Requirements:

 Benefits: This approach mandates disclosure of AI

involvement in content creation, including clear
indications of false and true information. It could aid in
combating deepfakes created by corporate entities, with
administrative fines potentially serving as a stronger
deterrent than criminal penalties. It aims to clearly indicate
when content has been produced using AI technology,
which is considered an important step.

 Concerns/Limitations: The concept of administrative

fines is unfamiliar in some countries, requiring extensive
cooperation and potentially constitutional changes. It is
uncertain if individuals will readily identify AI-labeled
materials as fake, as "AI-generated" doesn't necessarily
mean "fake," leading to confusion. There's a possibility that
people might even perceive AI-generated content as more
trustworthy due to natural biases. Lengthy transparency
statements may not be easily comprehensible, and true
algorithmic transparency may be impossible for regular
users or even experts to understand.

o 3. Imposition of Stricter Responsibilities on the Private

Sector:

 Benefits: Social media providers, mandated by laws like

the EU's DSA, can remove fake news and mitigate risks.
Some platforms already prohibit such content based on
internal policies. Obliging platforms to remove deepfakes
due to unlawful processing of personal data offers a
specific avenue.
  Concerns/Limitations: It's crucial to avoid viewing
private sector entities solely as problem-solvers, as their
content moderation can influence users'
perspectives and may serve to enhance visibility or
censor content diverging from their policies/interests.
Social media providers are primarily driven by financial
interests, which may conflict with safeguarding individual
rights (e.g., Meta's decision to end fact-checking in the
USA). This approach has limited efficiency and is not a
long-term solution. Imposing responsibilities on AI model
developers to prevent fake material could constrain
creative abilities and hinder commercial exploitation of
AI research.

 "Cognitive Liberty" as a Promising Solution: Cognitive liberty is

advocated as a promising solution because it addresses the core issue
of mental self-determination in the face of sophisticated manipulation.

o Definition and Relation to Deepfakes: It is broadly

understood as the right to mental self-determination,
ensuring an individual's sovereignty over their own mind. It
derives from the principle of individual autonomy, which means
being directed by one's authentic self, not externally imposed
influences. The sources advocate for a broader, negative
definition: the right not to be subjected to interference
with one’s cognitive processes. The creation and
dissemination of deepfakes directly fall within this definition
because they involve an intrusion into someone’s cognition
without direct brain intervention, modifying perception in
virtual environments and inhibiting the ability to compare
experiences.

o Implicit Recognition in Existing Human Rights: While there

is no international law directly recognizing cognitive liberty, it is
considered implicit in several existing human rights. These
include:

 The right to privacy.

 The right to freedom of opinion and expression.

 The right to freedom of thought, conscience, and religion.

  The right to education.

 The right to health.

 The right to participate in cultural life. The Charter of

Fundamental Rights of the European Union
recognizes the right to mental integrity (Article 3.1) as
"the right to be free from unwanted mental interference or
manipulation of a direct and forcible nature". Although
traditionally linked to biomedicine, torture, and human
dignity, this provision could benefit from further
specification to include threats from new technologies like
deepfakes.

o Challenges to Formal Acknowledgement and Effective

Enforcement:

 Law Enforcement Efficacy: The viability of protecting

cognitive liberty hinges on effective law enforcement,
which is often lacking in international law, especially during
conflicts.

 Lack of Consensus: Achieving consensus among nations,

particularly those reliant on information propaganda, to
enforce obligations protecting cognitive liberty under
international law remains uncertain.

 Extraterritoriality Challenges: While extraterritorial

human rights obligations (requiring states to prevent rights
violations originating from actions under their control, even
beyond borders) could apply to cross-border deepfake
dissemination, the practical execution of such
extraterritorial enforcement remains largely
hypothetical. It requires robust international cooperation,
which is challenging given differing approaches even
among allies (e.g., EU and USA regarding social media
accountability). Despite these challenges, acknowledging
cognitive liberty at the international level could facilitate a
new, universally acceptable standard for technologies
influencing decision-making.

 AI as a Solution to Deepfakes: Ironically, AI itself is seen as a key

defense against deepfakes.
 o How AI can be leveraged: Machine learning models can be
trained to detect AI-generated content by analyzing
inconsistencies in facial expressions, speech patterns, and
pixel anomalies. Researchers are exploring watermarking
methods and digital fingerprinting to authenticate real
content. AI-driven fact-checkers and detection tools need
continuous updates to keep pace with evolving deepfake
technologies. Investing in enhanced detection tools and fostering
collaboration among governments, academia, and the private
sector is crucial for developing effective solutions.

o Limitations of this Approach: Deepfake creators continuously

improve their techniques, making detection a constant
challenge. Current detection tools often struggle with real-
time identification, and disguised or low-quality deepfakes can
be particularly challenging to detect.

6. International Law and Future Directions

 Application of Geneva Conventions to Deepfakes in Wartime as

"Ruses": The sources argue that in the context of war, deepfakes
could be treated as ruses.

o Categorization as Ruses: A ruse is defined as a "trick of war

designed to deceive the adversary," often involving the
deliberate exposure of false information to enemy intelligence.
Examples include transmitting misleading messages, knowingly
permitting interception of false documents/plans, using enemy
wavelengths/codes for false instructions, or pretending to
communicate with non-existent reinforcements.

o Permissibility: Ruses are generally permissible under the

1899 and 1907 Hague Conventions and the 1949 and 1977
Geneva Conventions. Specifically, Article 37 of Additional
Protocol I to the Geneva Conventions explicitly recognizes
misinformation as a ruse. Experts' commentary on the Geneva
Conventions highlights that ruses can "permit a successful
operation with less loss of life than through the simple use of
force". Both Russia and Ukraine are parties to all four Geneva
Conventions and Additional Protocol I, obliging them to adhere to
its provisions, meaning deepfakes can be used as ruses within
 established limits, but never as perfidy (falsely claiming
international protection or breaching trust).

o Why this "narrow view...fails to capture the full potential

of deepfakes to cause harm": This categorization, while
legally plausible, is a narrow view that fails to account for
deepfakes' full potential for harm.

 Blurring Lawful Ruses and Prohibited Perfidy:

Deepfakes blur the line between lawful ruses and
prohibited acts of perfidy, which involve betraying trust.
For instance, deepfakes impersonating humanitarian
officials or fabricating evacuation calls could violate
International Humanitarian Law (IHL) if they intentionally
exploit civilian trust to cause harm.

 Violation of Precautions in Attack: Deepfakes

challenge the principle of precautions in attack (Article 57
of Additional Protocol I), which mandates directing attacks
solely at legitimate military objectives and minimizing
harm to civilians. A deepfake instructing civilians to
evacuate to areas later targeted could result in mass
casualties, violating this obligation.

 Outdated Legal Framework: The Geneva Conventions

were crafted when deepfakes were unimaginable. The
dangers they pose highlight an urgent need for
reevaluation of IHL to account for digital warfare
realities. This includes explicit prohibitions on creating or
disseminating deceptive media targeting civilian
populations and establishing accountability mechanisms.

 Recommendations for India's "Way Forward": To effectively tackle

deepfakes, India needs a comprehensive, multifaceted approach,
drawing on global comparisons and best practices.

o Deepfake-Specific Legislation:

 Strengthen Existing Laws: Amend the Information

Technology Act to specifically address the creation and
misuse of deepfakes, including provisions for penalties for
malicious use.
  New Legal Provisions: Introduce explicit legal
provisions under the Bharatiya Nyay Sanhita to
criminalize identity theft, extortion, defamation, and
harassment specifically through deepfakes. This will
provide clear legal boundaries for deepfake-related crimes
that existing laws do not fully cover.

 Mandate Transparency: Implement laws that mandate

watermarking of AI-generated content and require
transparency from creators regarding AI use.

 Platform Accountability: Impose liability on social media

platforms and service providers that fail to regulate
deepfake content.

o Technological Solutions:

 Advanced AI Detection Tools: Develop and deploy

advanced AI detection tools to detect deepfake content
in real-time. India can collaborate with global AI
leaders to build robust systems.

 Blockchain Systems: Implement blockchain systems

within media companies and government agencies to
ensure transparency and traceability of digital content.

 Content Monitoring Mandates: Mandate social media

platforms operating in India to implement robust content
monitoring tools for deepfake removal.

o Public Awareness and Education:

 Media Literacy Campaigns: Conduct comprehensive

media literacy campaigns to educate citizens on how to
recognize deepfakes, understand potential risks, and
critically evaluate digital content.

 Workshops and Training: Organize workshops and

training programs for professionals to identify and combat
deepfake misuse.

o Institutional and Collaborative Measures:

  Dedicated Regulatory Body: Establish a dedicated
regulatory body to oversee deepfake-related cases, AI
ethics, and digital rights.

 International Collaboration: India should collaborate

with global partners to develop an international legal
framework for combating cross-border deepfake crimes,
similar to those governing cybercrimes.

 Public-Private Collaboration: Foster a strong public-

private collaboration for developing robust policies and
ethical AI guidelines. By adopting a balanced legal
structure that addresses deepfake crimes while supporting
AI innovation, India can mitigate these threats effectively.

Here is a comprehensive list of important points about deepfakes, drawing

on the provided sources:

I. Understanding Deepfakes
  Deepfakes are generally understood as synthetic/artificial media
which is computer-generated and depicts a scenario that did not
actually occur, or depicts a person doing or saying something they did
not actually do or say. The term is a portmanteau of ‘deep learning’
and ‘fake’.

 They are digitally altered photos, videos, or audio that appear

incredibly realistic.

 The creation process involves feeding a computer program with

existing images, videos, or audio of a person, which then analyzes
facial expressions, movements, and voice patterns to generate new,
realistic content. This often utilizes Generative Adversarial
Networks (GANs), comprising a generator that produces fake content
and a discriminator that differentiates between authentic and fake
content.

 Deep synthesis is an umbrella term encompassing technologies like

deep learning and augmented reality, used to generate text, images,
audio, and video to construct virtual scenarios.

II. Harms and Impacts of Deepfakes

 Deepfakes pose significant challenges to cybersecurity by weakening

trust in digital media and creating security risks through misleading
information, impersonation, phishing attacks, and manipulation of
public opinion. They add another layer to cybersecurity threats,
impacting enterprises, national security (e.g., in geopolitical and
cyberwarfare situations), and society by misleading people.

 The harms caused to individuals can be significant, especially when

deepfakes are distributed without consent.

 They can lead to psychological harm, financial harm, and societal

harm.

o Individual Harms: Sextortion, defamation, intimidation,

bullying, undermining trust, extortion, identity theft, and
payment fraud. Examples include an employee tricked into
sending $25 million to fraudsters, a 77-year-old lecturer duped of
over £17k in a romance scam, and sexually explicit deepfakes of
classmates created by students.
 o Organizational/Group Harms: Stock-price manipulation, brand
image damage, and reputational damage. The CEO of Xiaomi
was a victim when AI-generated videos portraying him negatively
went viral.

o Societal Harms: News media manipulation, damage to

economic stability, damage to the justice system, damage to the
scientific system, erosion of trust, and damage to democracy
and national security.

 Deepfakes are overwhelmingly used to create non-consensual

pornography, often targeting women and girls. The Taylor Swift
deepfake incident sparked outrage and furthered conversations about
digital rights and regulation.

 They are used for spreading misinformation and manipulating

political events or speeches. Examples include a deepfake voice
mimicking President Biden used in robocalls during the 2024 New
Hampshire primary, and deepfake videos of politicians criticizing
opponents or campaigning for specific parties. The Russian-Ukrainian
war has seen deepfakes used to undermine morale, persuade the
international community, and discredit allies.

 In the courtroom context, deepfakes impact evidence authenticity,

witness credibility, and the integrity of the judicial process.
Even genuine evidence can be alleged to be false, requiring disproof.

 Deepfakes are a threat to automated speaker verification systems and

voice-controlled systems used in the Internet of Things, leading to
cybercrimes like fraud and extortion.

 They can also affect memory perception, leading to false memories

of events that never happened, and mislead jurors in legal cases.

III. Key Drivers of Deepfake Proliferation

 Advancing AI and ML technologies are becoming cheaper and

more accessible.

 Availability of datasets and computing power: Large, often freely

available, datasets of visual material, coupled with advancements in AI
and low-cost cloud computing, make it easy to create high-quality
deepfakes.
  Accessibility of high-quality algorithms and pre-trained
models: Openly published work and accessible code repositories are
misused, and pre-trained machine learning models can be reused with
malicious intent.

 5G connectivity enables higher-quality, portable videos with

increased high-speed bandwidth.

 Rise of 3D sensors in consumer electronics allows capturing 3D

information of scenes and scanning objects.

 The "cat and mouse game" between producers and detectors drives
increased quality in deepfake generation.

 Supply and demand platforms for deepfakes exist on the dark web.

 Commodification of tools (e.g., Dfaker, DeepFacelab, Faceswap) and

the emergence of "Deepfake as a service" companies at low cost.

IV. Challenges in Regulating and Detecting Deepfakes

 Lack of consensus on precise definition of a deepfake. It's difficult

to distinguish between simple media editing and AI-generated realistic
but fictional images.

 Technological complexity and rapid evolution: Deepfake

technology is constantly evolving, making it tough for regulators to
keep up with new developments and for detection tools to remain
effective.

 Difficulty of detection: Deepfakes are becoming increasingly difficult

to detect and are often almost indistinguishable from genuine content.
Detection algorithms are often short-lived and their reliability can be
significantly reduced by simple modifications in deepfake production
techniques.

 Anonymity of perpetrators and global nature of the internet:

Identifying creators and those who disseminate deepfakes is difficult,
especially when anonymous technologies or cross-border platforms are
used.

 Lack of universal standards and guidelines for creation and

dissemination.

 Balancing free speech and regulation: Overly broad laws risk

infringing on free speech and creative expression.
  Exclusive use of labels may not address the full extent of
harms and raises questions about target audience, display format, and
disclosure details.

 Copyright issues: Personality rights and IPR considerations are not

well-settled debates, making it unclear whether individuals can claim
deepfake creation violates their intellectual property interests.

 Deepfakes in end-to-end encrypted (E2EE) communication: A

notice-and-takedown approach or proactive monitoring is not feasible
without compromising privacy and security.

 Conflict with user rights: Proactive monitoring of content may

conflict with privacy and other rights-based considerations.

 Litigation costs: Parties may need to hire expensive digital forensics

experts to identify and debunk deepfakes.

 Erosion of trust in evidence: Even genuine evidence may be

doubted, leading to increased judicial skepticism and a higher burden
of proof. Jurors often lack the technical expertise to differentiate,
increasing the risk of prejudicial decisions.

V. International Policy and Legal Developments

A. United States

 Federal Level: Currently, there are no comprehensive federal laws

specifically regulating AI or deepfakes.

o NO FAKES Act 2023 (Proposed): A bipartisan bill prohibiting

the production of an indistinguishable digital replica of an
individual without consent, with exceptions for news, public
affairs, sports, documentaries, historical works, parody, or
criticism. Violations would be subject to civil action ($5000+ per
violation).

o US Malicious Deep Fake Prohibition Act of 2018

(Proposed): Aims to penalize creators of fake videos that
facilitate criminal or tortious conduct.

o National Defense Authorization Act (NDAA) of 2019,

Section 5709: Requires the Director of National Intelligence to
report on deepfake use by international governments and their
impact on national security.
 o DEEPFAKES Accountability Act of 2024 (Proposed):
Primarily focuses on preventing the use of deepfakes for
infringing sexual integrity.

o TAKE IT DOWN Act (April 2025): Addresses non-consensual

intimate imagery, including AI-generated deepfakes, providing a
mechanism for swift content removal and perpetrator
accountability.

 State Level: Some states have enacted laws due to the federal void.

o California:

 Assembly Bill 602 (AB 602): Provides the 'depicted

individual' with a cause of action against an individual for
creating and intentionally disclosing sexually explicit
material without consent, including digitally manipulated
content.

 Assembly Bill 730 (AB 730): Prohibited malicious

distribution of deceptive audio/visual media of a candidate
to injure reputation or deceive voters during election
periods, unless disclosed as manipulated. Its sunset period
ended Jan 1, 2023.

 Defending Democracy Act (2024): Targets deepfake

use during elections, requiring large platforms to remove
deepfakes within 120 days before an election and
extending disclosure requirements beyond that period
(effective Jan 1, 2025).

 SB970 (Feb 2024 proposed): Establishes standards for

identifying falsified evidence in court proceedings and
mandates the Judicial Council to develop rules by Jan 1,
2026, to assess AI-generated/manipulated evidence.

o Colorado AI Act: Signed into law in 2024, it is the first

comprehensive law in the US to impose obligations on
developers and deployers of AI systems, addressing "high-risk AI
systems" and deepfakes.

o Louisiana and Florida: Criminalized deepfakes showing minors

engaging in sexual conduct.
 o Oregon: Requires disclosure of synthetic media use in election
campaign communications.

o Tennessee and Mississippi: Legislated broadly against "the

unauthorised creation and distribution of a person’s photograph,
voice, or likeness".

o Existing laws: State-level defamation and privacy laws (e.g.,

California's Right of Publicity Law for commercial purposes) may
provide some recourse, but often require proving harm or specific
intent and may not cover non-commercial uses or images/videos
specifically.

B. United Kingdom

 Online Safety Act (OSA): Tackles deepfakes that are non-consensual

intimate images or film, making sharing a criminal offense (up to 2
years imprisonment), while creation is not regulated. The Act inserted
Section 66B into the Sexual Offences Act 2003, creating offences
related to sharing real and deepfake images, but not criminalizing
production alone.

 2025 Reforms ("Cracking Down on Deepfakes"): Section 135 of

the Data (Use and Access) Bill proposes new offences of "creating or
requesting the creation of a purported intimate image of an
adult" without consent or reasonable belief of consent. These offenses
are triable in Magistrates' Court, with up to six months custody and/or
an unlimited fine.

 UK GDPR and Data Protection Act 2018: A person can claim

misuse of personal data (including images) if processed without
consent.

 Defamation Act 2013: Allows suing for reputational damage, but

deepfakes' anonymity makes finding the responsible person difficult,
and requires proving "serious harm".

 The UK government is working on legislation to regulate AI models for

safety.

C. European Union

 Code of Practice on Disinformation + Digital Services Act

(DSA): DSA allows individuals to notify online platforms of illegal
content, though what's "illegal" isn't fully clarified. Recent discussions
 focus on requiring large online platforms to provide a visible indication
of malicious deepfakes. The Code of Practice on Disinformation,
enforced through DSA, requires platforms to tackle advertising with
disinformation and provide transparency on political advertising; failure
to comply can result in fines up to 6% of global turnover. The DSA aims
to reduce distribution of illegal content and protects consumer rights
online. It became effective February 17, 2024.

 EU AI Act (Proposed/Adopted 2024): Does not ban deepfakes

completely. Article 52(3) requires users of AI systems that generate
text/audio/visuals falsely appearing authentic to disclose in a timely,
clear, and visible manner that content is artificially
generated/manipulated. Exceptions include law enforcement
deepfake detection, and evidently creative, satirical, artistic works
where mere disclosure of use and copyrights is sufficient. Deepfakes
are classified as "limited risk" AI systems, in contrast to "high risk"
systems like medical AI or facial recognition. It defines deepfakes in
Article 3(60) as synthetic or manipulated content (audio or video)
deceptively appearing genuine and mirroring real individuals, places,
or events.

 GDPR: Plays a significant role; processing personal data (including

images) without consent could be a violation.

 EU Unfair Commercial Practices Directive (2005): Protects

consumer interests in behavioral advertising, prohibiting practices that
impair decision-making; could serve as a basis for further regulation.

D. China

 Regulations on the Administration of Deep Synthesis of

Internet Information Services (Regulations): Defines deep
synthesis technology broadly. The state internet information
department is responsible for oversight. Obligates deep synthesis
service (DSS) providers to verify real user identity, set up complaint
mechanisms. Article 17 requires DSS providers to make a
'conspicuous label' on generated/edited content, applicable in
specific scenarios (speech generation, image/video of virtual persons,
realistic immersive scenes). Prohibits bypassing/concealing the label.

 Imposes obligations on DSS providers and a general obligation on

individuals/organizations not to use deep synthesis services to produce
information that endangers national security/interests, harms national
 image/societal public interest, disturbs social/economic order, or harms
others' lawful rights.

 Enforced since January 2023, these "Deep Synthesis Provisions"

mandate enhanced data management, transparency, content
management, and technical security, ensuring personal information
protection. This approach prioritizes social stability and national
security, but raises concerns regarding freedom of information and
opinion.

E. Australia

 No specific laws about deepfakes.

 Criminal Code Amendment (Deepfake Sexual Material) Bill

(June 2024): Aims to criminalize sharing sexual material depicting
another person without their consent, applicable to both digitally
altered and unaltered material. Part of a wider agenda to tackle
gender-based violence.

 Defamation laws: Recognize that images (including digitally altered)

can be defamatory, but courts don't often grant injunctions for
removal.

F. France

 SREN Law (May 2024): Supplements Article 226-8 of the French

Criminal Code, explicitly prohibiting non-consensual sharing of
deepfake content unless its artificial generation is obvious.

 Updated Criminal Code to specifically criminalize sharing of non-

consensual pornographic deepfakes.

G. International Law

 In the context of war, deepfakes should be treated as ruses, which are

generally permissible under the 1899 and 1907 Hague Conventions
and the 1949 and 1977 Geneva Conventions. Both Russia and Ukraine
are parties to these.

 However, deepfakes blur the line between lawful ruses and prohibited
actions like perfidy (acts intended to betray trust).

 The Geneva Conventions were crafted before deepfakes existed,

highlighting an urgent need to reevaluate International Humanitarian
Law (IHL) to account for digital warfare realities, including explicit
 prohibitions on deceptive media targeting civilians and accountability
mechanisms.

VI. Industry Interventions and Technological Solutions

 Content Authenticity and Content Provenance: Initiatives like the

Content Authenticity Initiative and Coalition for Content Provenance
and Authentication propose cryptographically sealing and verifying
attribution information (where, how, when altered, by whom) from
creation to consumption. Microsoft offers "Content Credentials as a
Service" using C2PA digital watermarking.

 Watermarking:

o Meta's Stable Signature is an invisible watermarking technique

to distinguish AI-generated images, embedding the watermark
directly into the digital data for traceability.

o Google DeepMind's SynthID allows embedding digital

watermarks into AI-generated images or audio and scanning for
them to assess AI creation/alteration.

o Digitally watermarking audio-visual materials is a recommended

defense strategy.

 Algorithmic Detection: Automatic detection software relies on

detectable giveaways like speaker recognition, voice liveness
detection, facial recognition, facial feature analysis, temporal
inconsistencies, and visual artifacts.

o Microsoft Video Authenticator analyzes media to provide a

percentage chance of artificial manipulation.

o FakeCatcher (Intel partnership) detects synthetic videos by

analyzing subtle differences in skin color caused by the human
heartbeat.

o AI can be used to identify changes in voice and image that may

indicate deepfakes.

o AI-driven fact checkers and detection tools should be

continuously updated.

 Project Origin: A joint initiative (BBC, Microsoft) establishing an

engineering approach to synthetic media, providing digitally signed
 links for verifiable tracing back to the publisher and validation checks
against distribution changes.

 Platform Policy Changes:

o Meta: Updated political ads disclosure policy (Nov 2023)

requires advertisers to disclose AI-generated/altered content in
social issue, election-related, or political ads that misleads
viewers.

o YouTube: Will require creators to disclose realistic synthetic

media created with AI tools, with penalties for non-adherence
(content takedown, Partner Program suspension). Also allows
users to request removal of synthetic content with unauthorized
PII via their privacy complaint process.

o Google Play Store (Jan 31, 2024): Requires developers to

ensure apps comply with content policies (preventing restricted
content, child exploitation, deceptive behavior) and provide a
feature for flagging offensive content in generative AI apps.

 AI as a solution: Machine learning models can be trained to detect

generated content by analyzing inconsistencies in facial expressions,
speech patterns, and pixel anomalies. Blockchain technology can verify
content authenticity.

 User engagement and open communication: Policies should make

it easier for people to know the origin, creation method, and reliability
of information.

 Identity and Access Management (IAM) solutions & Multifactor

Authentication (MFA): Robust IAM practices including MFA provide
an extra layer of security.

 Red Teaming for AI systems: Systematic testing to identify

vulnerabilities and prevent misuse for deepfake creation.

 Continuous Monitoring and Threat Intelligence: Security

Operations Centers equipped with 24x7 monitoring and detection
capabilities.

 Project Origin: A joint initiative between media organizations like BBC

and Microsoft that aims to provide "digitally signed links that provide
verifiable tracing back of media content to the publisher" and
 "validation checks that the material was not changed during
distribution".

VII. Future Considerations and Recommendations

 Multi-dimensional approach: No single solution exists; a multi-

dimensional approach is required, with governments playing a crucial
role in policymaking.

 Transparency requirements: Focus on establishing transparency

requirements and awareness amongst end consumers about synthetic
media. This means clearly labeling deepfakes.

 Grievance redressal/reporting mechanism: Ensuring mechanisms

for reporting deepfakes.

 Internal policies and processes: Laying down internal policies for

identifying user accounts and enforcement against users
creating/sharing deepfakes without disclosure.

 Investment in technical interventions: More investment is needed

for better detection of new campaigns, identification of automated
content amplification, and analysis for action.

 Cognitive liberty: Protection against deepfakes should be sought

within the concept of cognitive liberty, understood as the right not to
have one’s mind interfered with by external influences.

 Criminalization of deepfakes: While proposed by some, it raises

concerns about infringing freedom of information and artistic
expression.

 Stricter responsibilities on the private sector: Imposing stricter

liability on service providers (e.g., social media platforms) to remove
fake news. This approach has limited efficiency and cannot be a long-
term solution.

 Cross-border cooperation: Essential due to the global nature of

deepfake dissemination.

 Public awareness and education campaigns: Launching

campaigns to educate voters about deepfake existence and impact,
and how to critically assess digital content. Media literacy programs
can empower individuals to recognize misinformation.
  Ethical AI development: Fostering the advancement of AI
technologies with ethical principles, prioritizing bias mitigation, privacy
safeguarding, and transparency.

 Regulatory frameworks must be flexible and adaptive to rapid

technological advancement.

 Corporate responsibility: Companies should implement robust

security measures and ethical guidelines for AI use, including self-
regulation and industry standards.

 Legal recourse for victims: Need for clear legal mechanisms for
victims to demand immediate takedowns or compensation.

 Adjustments to discovery rules in courts: To account for forensic

challenges posed by deepfakes.

 Education for judges and jurors: To understand deepfakes and

manage evidentiary challenges.

 Changes to the Rules of Evidence (US specific proposals):

Including a higher standard to prove authenticity, judges deciding on
authenticity rather than juries, and placing the burden on proponents
to show probative value.

VIII. Deepfakes in the Indian Context

 Legal Grey Area: India lacks a specific legal framework to combat

deepfakes directly, relying on outdated cyber and forgery laws.

 Existing Laws and their Limitations:

o IT Act, 2000 and IT Rules, 2021: Stipulate social media

intermediaries are responsible for promptly removing deepfake
content. Failure can result in imprisonment or fine.

 Section 66D: Penalizes impersonation through electronic

communication, applicable in financial fraud and extortion
cases involving deepfakes.

 Section 66E: Protects privacy by criminalizing

unauthorized capturing of images, applicable to deepfake
pornography and unauthorized use of private images.

 Section 67: Regulates obscene content, useful for non-

consensual deepfake pornography.
  Limitations: These provisions are not entirely sufficient;
they don't explicitly define deepfake-related offenses, were
not drafted with AI-generated content in mind, and don't
fully cover misinformation or political manipulation. They
also lack clear provisions for consent and harm mitigation.

o Indian Penal Code (IPC): Can be invoked for defamation and

cheating.

o Forgery laws (Sections 464 and 465 IPC): Criminalize

creation of false documentation with intent to deceive.

 Applicability: Can apply to deepfakes altering official

documents, financial fraud impersonations, false evidence
in legal matters, and identity theft.

 Limitations: Courts may struggle to decide if deepfakes

constitute "false documentation." They don't always result
in tangible harm, and are less suited for defamation,
political misinformation, or satire/parody which lack intent
to deceive.

 Electoral Challenges: Deepfakes pose a serious threat to free and

fair elections by spreading propaganda or discrediting opponents. The
Election Commission of India has issued guidelines against
misinformation, but enforcement is a challenge. Misinformation
manipulates public perception, enables character assassination,
suppresses voter turnout, and creates polarization.

 Cybersecurity Advisory: The Indian Computer Emergency Response

Team (CERT-In) issued a high-severity advisory, emphasizing caution in
sharing personal data, enabling privacy settings, verifying authenticity,
and looking for manipulation signs. For organizations, it recommended
watermarking, MFA, advanced detection tools, monitoring, and
strengthening legal/compliance frameworks.

 Liability of AI Developers and Platforms: A controversial issue;

arguments exist for holding tech companies accountable for providing
tools, content moderation failures, and negligence if deepfakes cause
harm. Section 79 of the IT Act provides intermediaries with a "safe
harbor" defense.

 Ownership of Face and Voice: Deepfakes raise critical questions

about data protection and consent. The Digital Data Protection Act
 2023 aims to safeguard personal data but doesn't explicitly cover
deepfake identity misuse.

 Proposed Solutions and Future Amendments for India:

o Dedicated legislation: India needs a specific legal framework

with provisions criminalizing malicious deepfake creations,
mandating transparency for AI-generated content, and holding
platforms accountable.

o Amendments: The IT Act needs amendments to specifically

address deepfake creation and misuse, including penalties.
Explicit legal provisions under the Bharatiya Nyay Sanhita to
criminalize identity theft, extortion, defamation, and harassment
through deepfakes are suggested.

o AI Detection Tools: Develop and deploy advanced AI detection

tools in real-time, potentially collaborating with global AI leaders.

o Blockchain systems: Media companies and government

agencies could implement blockchain for transparency and
traceability.

o Platform monitoring: Social media platforms operating in India

should be mandated to implement content monitoring tools to
remove deepfake content.

o Media literacy campaigns: Educate citizens on recognizing

deepfakes and potential risks.

o Regulatory body: Establishing a dedicated regulatory body for

AI ethics and digital rights may be necessary.

o International collaboration: India should collaborate with

global partners to address deepfake challenges effectively.

 Landmark Cases in India:

o Anil Kapoor vs. Simply Life India and Ors. (2023): Delhi
High Court granted an injunction against unauthorized use of Anil
Kapoor's persona via deepfakes, showing proactive judicial
stance.

o Rajat Sharma vs. Tamara and Ors. (2024): Delhi High Court
ordered removal of defamatory tweets and doctored videos. The
court balanced freedom of speech against the right to dignity.
o ANI vs. Open AI (2024): Indian news agency ANI sued Open AI
for using its published content without permission to train
ChatGPT, raising copyright concerns for AI-generated content.