Algorithmic Control and Platform Liability in the Streaming Era

This article is one of the winning entries from Academic IP Unveil 6.0 organised by Symbiosis Law School, Hyderabad.

Picture this, in Hyderabad, a 14-year-old boy simply sits in his room at night and watches Netflix. He is taking notes and formulating an action, he titles his plan “Mission Don”.

In August 2025 he breaks into the house of his neighbour by climbing a wall to steal a cricket bat. A 10-year-old girl is alone at home. She catches him. She screams. He does not run. He repeatedly stabs her almost 20 times.

Police found a diary named “Mission Don” which contained plans of theft motivated by Netflix programs, and a pair of bloodstained garments and the murder weapon.

Although the direct causation of this act of violence and streaming content has not been fully substantiated and proven, the incident reveals a regulatory fault line in the context of the intellectual property and intermediary liability law, which can no longer be overlooked. Netflix did not hand that boy a knife. But Netflix’s algorithm registered his engagement with violent content and served him more of it. It recommended. It personalised. It decided, through code, what this child’s screen looked like every time he opened the app. And then, after shaping his media environment through algorithmic recommendation, Netflix walked into court and said: We are just a neutral intermediary. We only host content. It is not our responsibility what people do with it.

That is the contradiction at the heart of Indian internet law right now.

The myth of neutral platforms

Consider a courier company. It selects a closed envelope and delivers it. It did not write the letter, chose who to write to and did not open the envelope.

This is the image underlying platform immunity law in India. Section 79, Information Technology Act, 2000 (IT Act) provides legal protection to intermediaries under three conditions: the platform must not have been the source of the transmission, chosen the recipient or altered the content. The immunity rewards one specific behaviour: Passivity. The Supreme Court reiterated this in Shreya Singhal v. Union of India¹, to make it clear that platforms can only lose immunity when a court order or a government directive is issued against them and not a private complaint.

This model reflected platforms in the 2000s, it does not describe Netflix, Instagram or YouTube in 2025. The experience that a user gets when he opens a streaming application is not that of an impartial library. It is a highly customised, refined, orchestrated space designed from the point of view of historical, demographic and behavioural patterns. It is an algorithm that controls visibility and makes things invisible. That is not a passive conduit. It is similar to editorial control — at a rate and scale that a human editor can never possibly handle, and with virtually no legal liability to its judgments.

Algorithmic copyright enforcement: They have lost the pretence

The clearest evidence of platform control emerges in copyright enforcement.

When a copyright holder’s money is at stake, platforms do not wait for court orders. They act instantly, automatically, and before any audience ever sees the content.

Using a database of audio and visual files submitted by copyright owners, Content ID identifies matches the moment a video is uploaded — scanning it before it reaches any viewer. Depending on the rights holder’s settings, a claim can result in the video being blocked, audio muted, revenue redirected, or access restricted by territory — any of these actions applied automatically, before any human reviews the content.

The scale demands attention. In 2024, a record 2.2 billion copyright claims were processed by Content ID on YouTube, of which over 99 per cent were automatically detected. No human adjudication. No court asked. No legal analysis of whether the use was transformative or protected as fair dealing. Since inception, Content ID has paid out $12 billion in advertising revenue to rights holders — $3 billion in 2024 alone, representing roughly 8.3 per cent of YouTube’s total advertising revenue that year and it runs entirely on code.

In one documented instance, a news channel uploaded public domain National Aeronautics and Space Administration (NASA) footage of a Mars rover and triggered claims against every other creator using that same footage — including against NASA itself. When a platform processes two billion enforcement decisions a year through pure code, individual accuracy becomes secondary to system design. The Electronic Frontier Foundation has documented the result: Creators abandon fair use entirely, endlessly editing lawful expressions just to satisfy the filter — because, as one creator with over a million subscribers put it, “the only thing that matters is, are you smarter than a robot”.

The point that cannot be argued around: When rights holder revenue is at stake, platforms exercise intensive, proactive, granular control over content — before any audience sees it, without any court order, through automated systems that override legal defences entirely. That is editorial governance.

The liability contradiction

According to the law, platforms are safeguarded as they are passive. According to the technology, platforms rank all content, personalise all feeds, scan all uploads against proprietary databases and make planet scale visibility decisions in real time.

Section 79 safe harbour is full of definitional loopholes and more of a performative than substantive compliance framework. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (IT Rules, 2021) only introduce quite procedural requirements — compliance reports, grievance mechanisms, paper content policies, and no requirement of proactive technical protection. These things are not necessary in the law. Yet they are already selectively used by platforms, when they have a financial interest in doing so.

Once a large music corporation tags a video, the action is automatic, instant and complete. Once the safety of the child is even possibly at risk, as a recommendation engine is prodding a teenager whose interaction activity indicates a worsening trend, the site once more becomes a passive intermediary. It is unable to keep track of everything. It is just a host. This is compounded by the IT Rules, they demand platforms to implement technology-based measures to safeguard children, but these same platforms can defend their passivity in court. The State cannot, at the same mandate algorhithmic control along with liberty to claim passivity. It is a contradiction platforms have been permitted to exploit, at the cost of their most vulnerable users.

Consequences and reform

Automated copyright enforcement operates as a structural suppression mechanism. Fewer than 1 per cent of Content ID claims were disputed in 2024 — yet that still amounts to 22 million disputes, and over 65 per cent were resolved in favour of the uploader. The system was wrong tens of millions of times. But most creators never dispute, because the process is designed to discourage challenge. Fair dealing — a statutory right under the Indian copyright law is effectively overridden by an algorithm that cannot perform the legal reasoning fair dealing requires.

Since 2019, 40 per cent of documentaries released on major Indian streaming platforms have been a true crime — a high-engagement category platforms have fed aggressively to Indian audiences. Research has shown that the international media, which idolise the anti-heroes depicted in TV series like Money Heist, can develop children’s love for criminal behaviour. The platform services economic benefits from both dynamics: Algorithmic enforcement silences creators; algorithmic amplification harms minors. Neither outcome attracts legal consequence.

Reform requires three moves. Firstly, stratify the definition of “intermediary”. A basic web host is genuinely closer to a courier. A feed personalisation platform, with a recommendation engine and a real-time visibility optic is categorically different. Active-versus-passive reasoning has already been used by Indian courts in relation to e-commerce and the same should be extended to streaming and social media. Secondly, replace the fiction of passivity with a demonstrable standard of care. The EU’s Digital Services Act (DSA)² is the working model requiring annual systemic risk assessments that factor in how recommender systems create harm, independently audited at the platform’s own expense, with penalties of up to 6 per cent of global annual revenue for non-compliance. India can build equivalent conditionality into Section 79 without abolishing safe harbour. Thirdly, make algorithm transparency mandatory. The DSA requires platforms to reveal the functionality of recommendation algorithms, as well as provide individuals with the option to stop feeds that are based on profiling without equivalent powers in India, courts decide liability in the dark, and every argument about platform responsibility is built on speculation.

Conclusion

A tragic incident involving two minors in Hyderabad illustrates the stakes of this debate. The algorithm that may have shaped the media environment around her killer is optimised for engagement, not safety. Nobody at Netflix made a conscious decision. But the code did — and the law currently has no answer for that.

Section 79 safe harbour is no longer viable in its present form. Platform with curative ranking, personalising, and enforcing platforms will not be able to claim protections available to entities that do not do any of those things. The responsibility of the law should follow power, not the old myth of a messenger. The law will no longer be able to afford to be ignorant of the decision-makers in a system where code is what determines what is seen.