Exclusive: Shadow AI in newsrooms - Should Indian media be worried?

Last month, Italian newspaper Il Foglio produced an AI-generated newspaper edition, marking what it describes as a historic moment for journalism. As part of a month-long initiative exploring the role of AI in media, the conservative-liberal daily produced a complete four-page broadsheet using AI tools. Il Foglio editor Claudio Cerasa explained that journalists’ involvement in the AI-produced edition would be minimal, limited to “asking questions [to an AI tool] and reading the answers.”

Well, journalists worldwide have welcomed AI with a childlike enthusiasm, as if a long-awaited, shiny toy had arrived. But, is everything as ‘shiny’ as it appears?

A grey area?

Countries the world over are realizing the need for a coherent AI policy. Recently, Indonesia's Press Council came up with a set of guidelines, aimed at ensuring the ethical and transparent use of AI in journalism.

Inputting sensitive data—such as source material, research, embargoed news, or copyrighted content—into Large Language Models (LLMs), especially through unapproved platforms, can do more harm than good. When done without the publication’s consent, this practice poses serious risks, including potential misuse of the data for AI model training, a serious bone of contention between media organisations and AI firms the world over.

A major concern is the lack of transparency around how some AI systems store and learn from user inputs. There’s a looming threat that such data could inadvertently become part of future AI model outputs, compromising confidentiality and intellectual property. This is precisely what journalists should be worried about when they input sensitive data into an AI system.

Globally, media organisations have already locked horns with AI companies over the unauthorized use of copyrighted content to train LLMs. In India, a legal battle is currently underway between OpenAI and ANI over alleged copyright infringement—reflecting the growing tension between content creators and AI developers.

Enter Shadow AI

As these legal and ethical debates unfold, a quieter but equally pressing challenge is emerging within newsrooms themselves: Shadow AI!

"Shadow AI" refers to the use of AI tools by employees for work without their company's knowledge or permission. Because many news organisations don't have clear AI policies, this practice could create significant problems for the future of their newsrooms.

According to a report by AI software company Trint, nearly half of the journalists surveyed admitted to using Gen AI tools that were neither approved nor provided by their organisations. The survey, which included reporters and editors from 29 newsrooms across the globe, highlights a growing trend of unsanctioned AI use in the media industry. In response, 64% of organisations said they plan to improve employee education, while 57% aim to introduce new policies on AI usage this year—underscoring the urgent need for clearer governance and training around AI in newsrooms.

The risks of shadow AI in Indian newsrooms are manifold; journalists using unapproved AI tools can lead to legal and ethical challenges, especially concerning data misuse by AI companies, copyright, editorial integrity, etc.

When sensitive data is input into unapproved AI systems, it may be used for model training and subsequently appear in the system's outputs. Furthermore, the lack of security in these systems could expose the data to the AI company, reviewers, or external parties.

Shadow AI is an intriguing phenomenon—one that, in my view, emerges when organisations fail to adopt a forward-thinking approach, points out Social Panga Co-founder Himanshu Arora.

“If a company restricts or ignores AI adoption, it’s not just hindering efficiency but also limiting the growth of both individuals and the organisation as a whole. AI isn’t something to shy away from or resist. When employees start using AI tools outside of the company’s mandated tech stack, Shadow AI emerges. Instead of viewing this as a threat, organisations should take a closer look and consider integrating AI ethically and as part of standard operating procedures. It’s like banning Google when the rest of the world is using it to complete tasks in a fraction of the time. Makes no sense, right?” Arora explains.

Indian newsrooms, like their global counterparts, are increasingly reliant on digital tools, says Meher Patel, Founder, Hector. The allure of AI's efficiency, from transcription to content generation, is undeniable. However, he adds, the unchecked adoption of unapproved AI tools introduces a host of risks:

• Data Breaches and Misuse:

○     Journalists often handle sensitive data, including confidential source material, unpublished documents, and copyrighted content. Inputting this data into unvetted AI models, particularly those hosted by third-party companies with unclear data policies, poses a significant risk of data breaches.

○     AI companies could potentially use this data to train their models, leading to unauthorized reproduction and distribution of sensitive information.

• Copyright Infringement:

○     Using AI tools to generate or manipulate content can inadvertently infringe upon existing copyrights. This becomes particularly problematic when dealing with multimedia content, where AI tools can easily remix or repurpose copyrighted material. The complex area of weather content, or data-generated content, could be very problematic.

• Editorial Integrity and Misinformation:

○     Unvetted AI tools may produce inaccurate or biased information, compromising the credibility of news reports. Without proper oversight, newsrooms risk disseminating misinformation, damaging their reputation and eroding public trust.

○     Inconsistencies in the use of AI tools across different departments or journalists can lead to varying editorial standards, creating confusion and undermining the organization's brand.

• Ethical Challenges:

○     AI tools can perpetuate existing biases, leading to discriminatory reporting.

○     The use of AI to generate deepfakes or manipulate video footage raises serious ethical concerns about the authenticity and integrity of news content.

• Legal Challenges:

○     With India enacting its own data protection laws, the storing and usage of personal information with un-vetted AI, can put media companies in very compromising legal positions.

○     Lack of transparency regarding where and how data is being processed can create legal challenges related to cross-border data flow.

Apoorva Kumar, CEO and Co-founder Inspeq AI, feels that Indian media must remain vigilant about shadow AI, as it can subtly manipulate public perception, distort information, and undermine trust in legitimate sources. To protect democracy, transparency, and societal integrity, it's crucial to implement strong AI safety standards, accountability frameworks, and proactive monitoring to mitigate potential risks, says Kumar.

Indian publishers, take note

Large publishers The Guardian, The Washington Post, The New York Times, and Gannett have developed formal AI policies, establishing ethical guidelines, and internal oversight mechanisms. These measures aim to protect company assets, intellectual property, and personnel.

A key aspect of these efforts involves educating newsroom employees about the risks of using unapproved AI tools. Publishers believe these policies and task forces are sufficient for responsible AI adoption. Some publishers, like The New York Times, are also whitelisting approved AI technologies.

However, smaller publishers with limited resources face significant challenges in implementing such comprehensive strategies.

Should Indian media organisations follow suit, setting up internal AI policies and mechanisms?

Experts feel that they should.

Meher Patel says that the experiences of leading international publishers like Gannett, The Guardian, The New York Times, and The Washington Post underscore the necessity of formal AI policies. These policies provide a framework for responsible AI adoption, mitigating risks and ensuring ethical compliance.

For Indian media companies, Patel adds, implementing similar policies is not a matter of choice but a crucial step towards safeguarding their future.

According to him, an AI policy and guidelines should address:

• Approved AI Tools:

○     Clearly define which AI tools are authorized for use and establish a process for evaluating and approving new tools.

• Data Handling and Security:

○     Establish strict guidelines for handling sensitive data, ensuring compliance with data privacy regulations.

○     Outline clear protocols for data storage, processing, and sharing.

• Copyright and Intellectual Property:

○     Provide clear guidelines on copyright compliance when using AI tools for content generation.

○     Establish procedures for obtaining necessary permissions and licenses.

• Editorial Standards and Accuracy:

○     Develop guidelines for verifying the accuracy of AI-generated content.

○     Establish clear protocols for human oversight and fact-checking.

• Ethical Considerations:

○     Address issues of bias, fairness, and transparency in AI usage.

○     Establish guidelines for the responsible use of AI in sensitive areas, such as deepfakes and manipulated media.

• Training and Education:

○     Provide journalists with training on the responsible use of AI tools.

○     Promote awareness of the potential risks and ethical challenges associated with AI.

• Auditing and accountability:

○     Establish a system for regular audits regarding AI usage.

○     Create accountability frameworks related to violations of the AI usage policy.

“By adopting comprehensive AI policies and guidelines, Indian media companies can harness the power of AI while mitigating its inherent risks. This proactive approach will be essential to maintaining public trust, upholding journalistic integrity, and ensuring the long-term sustainability of the industry in the age of AI,” he adds.

Himanshu Arora highlights that organizations are already introducing internal AI policies and frameworks to strike a balance—using AI for repetitive, redundant tasks while preserving human originality where it truly matters.

According to Arora, the key is to foster a culture where human-led creativity is empowered by AI-driven efficiency.

He suggests a few practical steps:

• Sanction a specific set of AI tools at the organizational level.
• Ensure leadership plays a proactive role in enabling access to these tools.
• Simultaneously, provide training on the risks of using unsanctioned AI tools.

This is particularly crucial in industries like journalism, where originality is the cornerstone. In such fields, the margin for error is razor-thin. We've already seen instances where AI-generated content slipped through editorial checks without proper verification—like CNET’s AI experiment or Amazon’s data leaks.

Closer to home, many Indian banks have taken a firm stance on AI usage, mandating that employees adhere strictly to an approved tech stack. Given the sensitivity of financial data, such measures are essential to ensure privacy and regulatory compliance.

However, Arora cautions that AI governance is a constantly evolving space. Organizations must take adaptive action—either integrate AI thoughtfully or risk being left behind.

“The faster you enable and regulate AI,” he says, “the quicker Shadow AI becomes a non-issue,” he concludes.

AI should definitely be viewed as a powerful tool to augment, rather than supplant, human reporting. By handling tasks like data analysis, editing, transcription, and routine reporting, AI allows journalists to focus on in-depth investigations and nuanced storytelling. This is particularly beneficial for resource-constrained local newsrooms, enabling them to maintain comprehensive coverage. AI also proves invaluable in investigative journalism, sifting through vast datasets to uncover hidden patterns. The future envisions a collaborative partnership between humans and AI, where technology enhances reporting and upholds journalistic ethics, ultimately serving the public interest. For such a collaborative approach to happen news rooms need to have robust internal policies in place.

(Watch this space—as Adgully will be bringing in journalists’ perspectives in the days ahead.)