#FactCheck : Old images of US sailors falsely linked to ongoing Iran tensions

The World Economic Forum reported that AI-generated misinformation and disinformation are the second most likely threat to present a material crisis on a global scale in 2024 at 53% (Sept. 2023). Artificial intelligence is automating the creation of fake news at a rate disproportionate to its fact-checking. It is spurring an explosion of web content mimicking factual articles that instead disseminate false information about grave themes such as elections, wars and natural disasters. 

According to a report by the Centre for the Study of Democratic Institutions, a Canadian think tank, the most prevalent effect of Generative AI is the ability to flood the information ecosystem with misleading and factually-incorrect content. As reported by Democracy Reporting International during the 2024 elections of the European Union, Google's Gemini, OpenAI’s ChatGPT 3.5 and 4.0, and Microsoft’s AI interface ‘CoPilot’ were inaccurate one-third of the time when engaged for any queries regarding the election data.   Therefore, a need for an innovative regulatory approach like regulatory sandboxes which can address these challenges while encouraging responsible AI innovation is desired.

What Is AI-driven Misinformation?

False or misleading information created, amplified, or spread using artificial intelligence technologies is AI-driven misinformation. Machine learning models are leveraged to automate and scale the creation of false and deceptive content. Some examples are deep fakes, AI-generated news articles, and bots that amplify false narratives on social media.

The biggest challenge is in the detection and management of AI-driven misinformation. It is difficult to distinguish AI-generated content from authentic content, especially as these technologies advance rapidly. 

AI-driven misinformation can influence elections, public health, and social stability by spreading false or misleading information. While public adoption of the technology has undoubtedly been rapid, it is yet to achieve true acceptance and actually fulfill its potential in a positive manner because there is widespread cynicism about the technology - and rightly so. The general public sentiment about AI is laced with concern and doubt regarding the technology’s trustworthiness, mainly due to the absence of a regulatory framework maturing on par with the technological development. 

Regulatory Sandboxes: An Overview

Regulatory sandboxes refer to regulatory tools that allow businesses to test and experiment with innovative products, services or businesses under the supervision of a regulator for a limited period. They engage by creating a controlled environment where regulators allow businesses to test new technologies or business models with relaxed regulations.

Regulatory sandboxes have been in use for many industries and the most recent example is their use in sectors like fintech, such as the UK’s Financial Conduct Authority sandbox. These models have been known to encourage innovation while allowing regulators to understand emerging risks. Lessons from the fintech sector show that the benefits of regulatory sandboxes include facilitating firm financing and market entry and increasing speed-to-market by reducing administrative and transaction costs. For regulators, testing in sandboxes informs policy-making and regulatory processes. Looking at the success in the fintech industry, regulatory sandboxes could be adapted to AI, particularly for overseeing technologies that have the potential to generate or spread misinformation.

The Role of Regulatory Sandboxes in Addressing AI Misinformation

Regulatory sandboxes can be used to test AI tools designed to identify or flag misinformation without the risks associated with immediate, wide-scale implementation. Stakeholders like AI developers, social media platforms, and regulators work in collaboration within the sandbox to refine the detection algorithms and evaluate their effectiveness as content moderation tools.

These sandboxes can help balance the need for innovation in AI and the necessity of protecting the public from harmful misinformation. They allow the creation of a flexible and adaptive framework capable of evolving with technological advancements and fostering transparency between AI developers and regulators. This would lead to more informed policymaking and building public trust in AI applications.

CyberPeace Policy Recommendations 

Regulatory sandboxes offer a mechanism to predict solutions that will help to regulate the misinformation that AI tech creates. Some policy recommendations are as follows: 

1. Create guidelines for a global standard for including regulatory sandboxes that can be adapted locally and are useful in ensuring consistency in tackling AI-driven misinformation.
2. Regulators can propose to offer incentives to companies that participate in sandboxes. This would encourage innovation in developing anti-misinformation tools, which could include tax breaks or grants.
3. Awareness campaigns can help in educating the public about the risks of AI-driven misinformation and the role of regulatory sandboxes can help manage public expectations.
4. Periodic and regular reviews and updates to the sandbox frameworks should be conducted to keep pace with advancements in AI technology and emerging forms of misinformation should be emphasized.

‍

Conclusion and the Challenges for Regulatory Frameworks

Regulatory sandboxes offer a promising pathway to counter the challenges that AI-driven misinformation poses while fostering innovation. By providing a controlled environment for testing new AI tools, these sandboxes can help refine technologies aimed at detecting and mitigating false information. This approach ensures that AI development aligns with societal needs and regulatory standards, fostering greater trust and transparency. With the right support and ongoing adaptations, regulatory sandboxes can become vital in countering the spread of AI-generated misinformation, paving the way for a more secure and informed digital ecosystem.

References

• https://www.thehindu.com/sci-tech/technology/on-the-importance-of-regulatory-sandboxes-in-artificial-intelligence/article68176084.ece
• https://www.oecd.org/en/publications/regulatory-sandboxes-in-artificial-intelligence_8f80a0e6-en.html
• https://www.weforum.org/publications/global-risks-report-2024/
• https://democracy-reporting.org/en/office/global/publications/chatbot-audit#Conclusions

‍

Executive Summary

‍
A video is being shared on social media claiming that a female tourist from Delhi fell into a deep gorge during ziplining in Karnaprayag. In the viral video, a woman dressed in bright pink and yellow clothes is seen ziplining amidst high mountains, when suddenly the zipline wire breaks and she falls from a significant height into a deep gorge. Social media users are sharing this video, presenting it as a real incident. Research by the CyberPeace Research Wing  revealed that the claim of a Delhi woman tourist's death during a zipline accident in Karnaprayag is baseless. The viral video is AI-generated.

‍

Claim

‍

According to the claim, this accident took place in Karnaprayag, Uttarakhand, where a female tourist from Delhi became a victim of this horrific mishap. Several social media users are claiming that the woman died in this accident.

https://www.facebook.com/reel/26108544868822432

https://archive.is/CeNOV 

‍

Fact Check

‍

To investigate the video viral as a zipline accident in Karnaprayag, we conducted a reverse search of its keyframes. During this, we found the video uploaded as a Short on a Pakistani YouTube channel, @Zoyaqueen-w2t, on June 2, 2026. Hashtags like '#funnyshorts' were used in the caption of this video. Scanning this account revealed that several other fictional videos related to similar zipline accidents have also been uploaded here. Apart from this, the video was also shared with funny hashtags in some social media posts from May 2026.

https://www.youtube.com/shorts/Gzha_J7Fqv0

‍

‍

Following this, searching with relevant keywords yielded no credible media reports regarding any such zipline accident in Karnaprayag, Uttarakhand, in recent times. Subsequently, we scanned the viral video of the alleged zipline accident using the AI detection tool 'Hive Moderation'. During the analysis, the tool classified the video as highly likely to be AI-generated, with a score of 99.2%.

‍

‍

‍

Conclusion

‍

From the evidence gathered in our research , it is clear that the claim regarding the death of a female tourist from Delhi during a zipline accident in Karnaprayag is baseless. The viral video is not of a real incident, but is AI-generated.

‍

The Emerging Landscape of AI-Enabled Sign Language Technologies 

‍

Consumer technology has been moving in a single, steady direction for decades: machines are getting increasingly sensitive to human speech. The phone transcribes what we say into it. An algorithm responds to a question we ask. With just one tap, we can translate across languages. However, sign language, one of the most essential forms of human expression, has largely escaped this change for millions of Deaf and hard-of-hearing people. At last, that omission may finally be narrowing. 

For the first time, sign language recognition is now widely available in consumer goods thanks to Google DeepMind’s massively multilingual sign language-to-text model. Starting with American Sign Language to English, the technology currently powers sign-to-text dictation within Gboard and Live Transcribe on Pixel 11. Additional devices and languages are promised. In addition to using Live Transcribe to sign during live conversations, users can sign anywhere they would normally type, such as while conducting a web search, writing a message or interacting with an AI assistant. However, this development’s importance goes far beyond a single accessibility function. 

‍

Reimagining How We connect with Technology

The most noteworthy is a conceptual change, sign language is starting to be recognised as a valid interface for human-computer interaction in and of itself , rather than as a modality that technology can accept. Because sign languages are not spoken languages that are represented by hand, this distinction is important. The hands, arms, torso, head and facial expressions all simultaneously convey meaning in these independent natural languages, which have their own grammar, vocabulary and syntax. Compared to traditional voice recognition, this presents a far more complex computing task. 

‍

In terms of architecture, the model does not keep raw video instead, it processes pose landmark sequences. The original footage may be destroyed while an on-device mechanism tracks locations on the signer’s body and transmits only geometric coordinates for translation. Instead than using intermediate “gloss” representations, which sometimes lose the spatial and non-manual components crucial to meaning, translation happens directly. Accessibility and privacy meet at this point, a technology designed to grant independence shouldn’t require the surrender of personal biometric information in return. 

‍

The Indian question is larger than ASL

A more significant concern for India is raised by this development, whose sign language will artificial intelligence eventually comprehend? It is not possible to import an ASL-to-English model and claim it to be a solution. The linguistic architecture, communities and regional variations of Indian Sign Language are unique. ISL recognition and its translation into Hindi, Telugu and Bengali are the subject of an expanding amount of study yet this same research openly highlights the shortcomings of existing systems including limited vocabularies, isolated word recognition and noticeable sensitivity to individual signing style. 

This is not a coincidental distinction. Benchmark accuracy alone cannot be used to gauge inclusive AI; instead, it must be effective under typical circumstances for a variety of individuals, geographical locations and sign languages. It is clear from research on low resource sign languages that over three hundred sign languages are still woefully under-resourced and under-documented. A growing body of research supports signer-adaptive modelling, privacy preserving representations, community co-design and dialectical variety preservation. Therefore, making data collecting, engagement and design more truly inclusive may be the next real advancement rather than further scaling models. 

‍

A Legal Architecture already in place

India’s statutory framework offers a firm foundation for this trajectory. The Rights of Persons with Disabilities Act, 2016 defines universal design broadly enough to include cutting edge technologies and assistive devices. It is based on the ideals of equality, dignity, participation and accessibility. Information and communication technology access is specifically covered by Section 42, which requires captioning, sign language interpretation, accessible electronic content and universal design in common electronic products. In this context, accessible AI is an issue of statutory rights rather than technological generosity. This stance is supported by the UN Convention on the Rights of Persons with Disabilities, which addresses accessibility in Article 9 and information access and freedom of speech in Article 21. As a result, the central policy topic is changing  from whether technology should be made accessible to how accessibility should be incorporated from the start.

‍

Beyond sign-to-text

Beyond Transcription, Google has expressed aspirations for more sign languages, sign language production and expanded AI capabilities. Future architecture could be imagined as running along a continuous circuit that connects sign, text, speech and AI in both directions rather than just from sign to text. A consumer could deal with a bank without completely relying on an interpreter; a student could learn in her favourite language; or someone could sign a question to an assistant and could get an answer in generated sign language. However , there are still important unanswered questions about this future, such as who owns the data used to train these systems, how signers’ meaningful consent is obtained, how systematic misrecognition of specific communities is prevented and who is responsible when translation fails in an emergency, legal or medical setting. 

‍

The Real Measure of Inclusion

The mere fact that a machine has discovered something humans have long understood makes it easy to characterise innovations like these as breakthroughs. The most accurate way to put it is that technology becomes inclusive when individuals can use it without changing who they are or how they interact, not when it acknowledges more human behaviours. This means that India must continue to invest not only in models but also in Indian Sign Language databases, community-led research, accessibility standards and the meaningful involvement of the Deaf and hard-of-hearing populations in both design and evaluation. The ability of a system to detect a hand gesture will not define the future of accessible AI. 

‍

References

• https://aclanthology.org/2025.wslp-main.5/ 
• The Rights of Persons with Disabilities Act 2016 (Act No 49 of 2016), s 42.  
• Convention on the Rights of Persons with Disabilities (adopted 13 December 2006, entered into force 3 May 2008) 2515 UNTS 3, art 9.

‍