#FactCheck: Old clip of Greenland tsunami depicts as tsunami in Japan

Introduction
‍

Generative AI, particularly deepfake technology, poses significant risks to security in the financial sector. Deepfake technology can convincingly mimic voices, create lip-sync videos, execute face swaps, and carry out other types of impersonation through tools like DALL-E, Midjourney, Respeecher, Murf, etc, which are now widely accessible and have been misused for fraud. For example, in 2024, cybercriminals in Hong Kong used deepfake technology to impersonate the Chief Financial Officer of a company, defrauding it of $25 million. Surveys, including Regula’s Deepfake Trends 2024 and Sumsub reports, highlight financial services as the most targeted sector for deepfake-induced fraud.

Deepfake Technology and Its Risks to Financial Systems

India’s financial ecosystem, including banks, NBFCs, and fintech companies, is leveraging technology to enhance access to credit for households and MSMEs. The country is a leader in global real-time payments and its digital economy comprises 10% of its GDP.  However, it faces unique cybersecurity challenges. According to the RBI’s 2023-24 Currency and Finance report, banks cite cybersecurity threats, legacy systems, and low customer digital literacy as major hurdles in digital adoption. Deepfake technology intensifies risks like:

• Social Engineering Attacks: Information security breaches through phishing, vishing, etc. become more convincing with deepfake imagery and audio.
• Bypassing Authentication Protocols: Deepfake audio or images may circumvent voice and image-based authentication systems, exposing sensitive data.
• Market Manipulation: Misleading deepfake content making false claims and endorsements can harm investor trust and damage stock market performance.
• Business Email Compromise Scams: Deepfake audio can mimic the voice of a real person with authority in the organization to falsely authorize payments.
• Evolving Deception Techniques: The usage of AI will allow cybercriminals to deploy malware that can adapt in real-time to carry out phishing attacks and inundate targets with increased speed and variations. Legacy security frameworks are not suited to countering automated attacks at such a scale.

Existing Frameworks and Gaps

In 2016, the RBI introduced cybersecurity guidelines for banks, neo-banking, lending, and non-banking financial institutions, focusing on resilience measures like Board-level policies, baseline security standards, data leak prevention, running penetration tests, and mandating Cybersecurity Operations Centres (C-SOCs). It also mandated incident reporting to the RBI for cyber events. Similarly, SEBI’s Cybersecurity and Cyber Resilience Framework (CSCRF) applies to regulated entities (REs) like stock brokers, mutual funds, KYC agencies, etc., requiring policies, risk management frameworks, and third-party assessments of cyber resilience measures. While both frameworks are comprehensive, they require updates addressing emerging threats from generative AI-driven cyber fraud.

Cyberpeace Recommendations 

• AI Cybersecurity to Counter AI Cybercrime: AI-generated attacks can be designed to overwhelm with their speed and scale. Cybercriminals increasingly exploit platforms like LinkedIn, Microsoft Teams, and Messenger, to target people.  More and more organizations of all sizes will have to use AI-based cybersecurity for detection and response since generative AI is becoming increasingly essential in combating hackers and breaches.
• Enhancing Multi-factor Authentication (MFA):  With improving image and voice-generation/manipulation technologies, enhanced authentication measures such as token-based authentication or other hardware-based measures, abnormal behaviour detection, multi-device push notifications, geolocation verifications, etc. can be used to improve prevention strategies.  New targeted technological solutions for content-driven authentication can also be implemented.
• Addressing Third-Party Vulnerabilities: Financial institutions often outsource operations to vendors that may not follow the same cybersecurity protocols, which can introduce vulnerabilities. Ensuring all parties follow standardized protocols can address these gaps.
• Protecting Senior Professionals: Senior-level and high-profile individuals at organizations are at a greater risk of being imitated or impersonated since they hold higher authority over decision-making and have greater access  to sensitive information. Protecting their identity metrics through technological interventions is of utmost importance.
• Advanced Employee Training:  To build organizational resilience, employees must be trained to understand how generative and emerging technologies work. A well-trained workforce can significantly lower the likelihood of successful human-focused human-focused cyberattacks like phishing and impersonation.
• Financial Support to Smaller Institutions: Smaller institutions may not have the resources to invest in robust long-term cybersecurity solutions and upgrades. They require financial and technological support from the government to meet requisite standards.

Conclusion

According to The India Cyber Threat Report 2025 by the Data Security Council of India (DSCI) and Seqrite, deepfake-enabled cyberattacks, especially in the finance and healthcare sectors, are set to increase in 2025. This has the potential to disrupt services, steal sensitive data, and exploit geopolitical tensions, presenting a significant risk to the critical infrastructure of India. 

As the threat landscape changes,  institutions will have to continue to embrace AI and Machine Learning (ML) for threat detection and response. The financial sector must prioritize robust cybersecurity strategies, participate in regulation-framing procedures, adopt AI-based solutions, and enhance workforce training, to safeguard against AI-enabled fraud. Collaborative efforts among policymakers, financial institutions, and technology providers will be essential to strengthen defenses.

‍

Sources 

• https://sumsub.com/newsroom/deepfake-cases-surge-in-countries-holding-2024-elections-sumsub-research-shows/ 
• https://www.globenewswire.com/news-release/2024/10/31/2972565/0/en/Deepfake-Fraud-Costs-the-Financial-Sector-an-Average-of-600-000-for-Each-Company-Regula-s-Survey-Shows.html 
• https://www.sipa.columbia.edu/sites/default/files/2023-05/For%20Publication_BOfA_PollardCartier.pdf 
• https://edition.cnn.com/2024/02/04/asia/deepfake-cfo-scam-hong-kong-intl-hnk/index.html 
• https://www.rbi.org.in/Commonman/English/scripts/Notification.aspx?Id=1721 
• https://elplaw.in/leadership/cybersecurity-and-cyber-resilience-framework-for-sebi-regulated-entities/ 
• https://economictimes.indiatimes.com/tech/artificial-intelligence/ai-driven-deepfake-enabled-cyberattacks-to-rise-in-2025-healthcarefinance-sectors-at-risk-report/articleshow/115976846.cms?from=mdr 

‍

Introduction 

The link between social media and misinformation is undeniable. Misinformation, particularly the kind that evokes emotion, spreads like wildfire on social media and has serious consequences, like undermining democratic processes, discrediting science, and promulgating hateful discourses which may incite physical violence. If left unchecked, misinformation propagated through social media has the potential to incite social disorder, as seen in countless ethnic clashes worldwide.  This is why social media platforms have been under growing pressure to combat misinformation and have been developing models such as fact-checking services and community notes to check its spread. This article explores the pros and cons of the models and evaluates their broader implications for online information integrity.

How the Models Work 

1. Third-Party Fact-Checking Model (formerly used by Meta) Meta initiated this program in 2016 after claims of extraterritorial election tampering through dis/misinformation on its platforms. It entered partnerships with third-party organizations like AFP and specialist sites like Lead Stories and PolitiFact, which are certified by the International Fact-Checking Network (IFCN) for meeting neutrality, independence, and editorial quality standards. These fact-checkers identify misleading claims that go viral on platforms and publish verified articles on their websites, providing correct information. They also submit this to Meta through an interface, which may link the fact-checked article to the social media post that contains factually incorrect claims. The post then gets flagged for false or misleading content, and a link to the article appears under the post for users to refer to. This content will be demoted in the platform algorithm, though not removed entirely unless it violates Community Standards. However, in January 2025, Meta announced it was scrapping this program and beginning to test X’s Community Notes Model in the USA, before rolling it out in the rest of the world. It alleges that the independent fact-checking model is riddled with personal biases, lacks transparency in decision-making, and has evolved into a censoring tool. 
2. Community Notes Model ( Used by X and being tested by Meta): This model relies on crowdsourced contributors who can sign up for the program, write contextual notes on posts and rate the notes made by other users on X. The platform uses a bridging algorithm to display those notes publicly, which receive cross-ideological consensus from voters across the political spectrum. It does this by boosting those notes that receive support despite the political leaning of the voters, which it measures through their engagements with previous notes. The benefit of this system is that it is less likely for biases to creep into the flagging mechanism. Further, the process is relatively more transparent than an independent fact-checking mechanism since all Community Notes contributions are publicly available for inspection, and the ranking algorithm can be accessed by anyone, allowing for external evaluation of the system by anyone. 

CyberPeace Insights

Meta’s uptake of a crowdsourced model signals social media’s shift toward decentralized content moderation, giving users more influence in what gets flagged and why. However, the model’s reliance on diverse agreements can be a time-consuming process. A study (by Wirtschafter & Majumder, 2023) shows that only about 12.5 per cent of all submitted notes are seen by the public, making most misleading content go unchecked. Further, many notes on divisive issues like politics and elections may not see the light of day since reaching a consensus on such topics is hard. This means that many misleading posts may not be publicly flagged at all, thereby hindering risk mitigation efforts. This casts aspersions on the model’s ability to check the virality of posts which can have adverse societal impacts, especially on vulnerable communities. On the other hand, the fact-checking model suffers from a lack of transparency, which has damaged user trust and led to allegations of bias. 

Since both models have their advantages and disadvantages, the future of misinformation control will require a hybrid approach. Data accuracy and polarization through social media are issues bigger than an exclusive tool or model can effectively handle. Thus, platforms can combine expert validation with crowdsourced input to allow for accuracy, transparency, and scalability. 

Conclusion 

Meta’s shift to a crowdsourced model of fact-checking is likely to have bigger implications on public discourse since social media platforms hold immense power in terms of how their policies affect politics, the economy, and societal relations at large.  This change comes against the background of sweeping cost-cutting in the tech industry, political changes in the USA and abroad, and increasing attempts to make Big Tech platforms more accountable in jurisdictions like the EU and Australia, which are known for their welfare-oriented policies. These co-occurring contestations are likely to inform the direction the development of misinformation-countering tactics will take.  Until then, the crowdsourcing model is still in development, and its efficacy is yet to be seen, especially regarding polarizing topics. 

References 

• https://www.cyberpeace.org/resources/blogs/new-youtube-notes-feature-to-help-users-add-context-to-videos
• https://en-gb.facebook.com/business/help/315131736305613?id=673052479947730
• http://techxplore.com/news/2025-01-meta-fact.html 
• https://about.fb.com/news/2025/01/meta-more-speech-fewer-mistakes/ 
• https://communitynotes.x.com/guide/en/about/introduction 
• https://blogs.lse.ac.uk/impactofsocialsciences/2025/01/14/do-community-notes-work/?utm_source=chatgpt.com 
• https://www.techpolicy.press/community-notes-and-its-narrow-understanding-of-disinformation/ 
• https://www.rstreet.org/commentary/metas-shift-to-community-notes-model-proves-that-we-can-fix-big-problems-without-big-government/ 
• https://tsjournal.org/index.php/jots/article/view/139/57 

‍

EXECUTIVE SUMMARY:

A viral video is surfacing claiming to capture an aerial view of Mount Kailash that has breathtaking scenery apparently providing a rare real-life shot of Tibet's sacred mountain. Its authenticity was investigated, and authenticity versus digitally manipulative features were analyzed.

CLAIMS:

The viral video claims to reveal the real aerial shot of Mount Kailash, as if exposing us to the natural beauty of such a hallowed mountain. The video was circulated widely in social media, with users crediting it to be the actual footage of Mount Kailash.

FACTS:

The viral video that was circulated through social media was not real footage of Mount Kailash. The reverse image search revealed that it is an AI-generated video created by Sonam and Namgyal, two Tibet based graphic artists on Midjourney. The advanced digital techniques used helped to provide a realistic lifelike scene in the video.

No media or geographical source has reported or published the video as authentic footage of Mount Kailash. Besides, several visual aspects, including lighting and environmental features, indicate that it is computer-generated.

For further verification, we used Hive Moderation, a deep fake detection tool to conclude whether the video is AI-Generated or Real. It was found to be AI generated.

CONCLUSION:

The viral video claiming to show an aerial view of Mount Kailash is an AI-manipulated creation, not authentic footage of the sacred mountain. This incident highlights the growing influence of AI and CGI in creating realistic but misleading content, emphasizing the need for viewers to verify such visuals through trusted sources before sharing.

• Claim: Digitally Morphed Video of Mt. Kailash, Showcasing Stunning White Clouds
• Claimed On: X (Formerly Known As Twitter), Instagram
• Fact Check: AI-Generated (Checked using Hive Moderation).