Can Fashion Become A Form Of Resistance To AI Surveillance?

What if the clothes you wear could confuse the surveillance camera watching you? Not by hiding your face, but by making it harder for an AI system to recognise your body in the first place. As surveillance technologies become more sophisticated, some designers are looking to an unlikely place for resistance: the wardrobe.

This is the idea behind adversarial fashion, an emerging intersection of textile design and machine learning. Brands like Cap_able, AntiAI Clothing, Yelo Pomelo, Oogaly and RiotAssembly are experimenting with patterns designed to disrupt how computer vision systems interpret images. All of these initiatives are based outside India.

But confusing a camera is not quite as simple as wearing a loud print. Cap_able founder Rachele Didero points out that person detection and facial recognition are two different stages of the surveillance process. 

A system may first need to detect and locate a person before attempting to identify them. Cap_able's garments target this initial stage, potentially disrupting what follows. 

There is, however, a lot more science behind these patterns than meets the eye. Their effectiveness depends not just on the design, but also on the material, the body wearing it, movement, lighting and distance from the camera. 

As Didero told Decode, "The textile is not simply a surface onto which a digital pattern is printed; its physical properties are part of the design process itself."

And even a pattern that works today may not fool tomorrow's algorithms. Different machine-learning models interpret images differently, and systems evolve with new training data. Didero therefore sees adversarial fashion as "an ongoing area of research rather than a permanent technical solution".

There is a reason this conversation matters in India too. Delhi Police recently used extensive facial recognition to identify people at the Jantar Mantar protests in July. Surveillance cameras and AI-enabled vans are also becoming part of how the police monitor public spaces. For people walking past these cameras, there is often little clarity on how their faces are being scanned, what the technology makes of them or where that information goes.

The research behind adversarial fashion does not offer a ready answer to these concerns, but it does raise an interesting question about the technology itself. How much should we trust machines to see us accurately? 

"Machine perception is not neutral or infallible," Didero says. And if machines are getting better at watching us, perhaps we can get a little better at fooling them too.

Here are five stories that caught our attention this week.

Content Takedowns On Meta Platforms Rise 5.5% In India 

Meta took action against 20.21 million pieces of content across Facebook, Instagram and Threads in India in September 2026, a 5.5% increase from 19.15 million in August. Instagram accounted for the largest share, with 15.12 million pieces of content actioned, followed by Facebook at 4.58 million. The transparency report also recorded a decline in proactive detection of bullying, harassment and hate speech on both platforms, while orders received from Grievance Appellate Committees more than doubled to 926. These figures are estimates based on users Meta believes are in India and reflect enforcement of its platform policies, not content restricted in response to government or court orders.

MIB Wants Digital Platforms To Get Their Newsroom Practices In Order

The Ministry of Information and Broadcasting (MIB) has issued a nine-point advisory to OTT platforms, digital news publishers and online intermediaries, citing instances of non-compliance with the IT Act and its rules. The advisory asks publishers to verify AI-generated and altered content, disclose the use of synthetic media, and provide the original date and location of archival visuals. It also flags coverage of anti-terror operations, communal issues and crime, and calls for stronger internal editorial checks. However, it does not identify the publishers or platforms allegedly in violation or specify the instances of non-compliance.

Bombay High Court Sets Aside Customs Order Over Fake AI-Generated Judgments

The Bombay High Court has set aside an order passed by the Commissioner of Customs after a petitioner flagged that it cited two Supreme Court judgments that did not exist and appeared to have been generated using AI. The petitioner argued that the Commissioner had relied on these unverified judgments without independently checking their authenticity. It also said that its request for copies of the cited judgments had gone unanswered. The High Court set aside the order with the consent of both parties and directed the Commissioner to reconsider the matter. The court order does not name the two allegedly fabricated judgments.

Google Faces Lawsuit Over AI Chats Allegedly Encouraging Suicide 

The parents of a 13-year-old girl in the US have sued Google, alleging that its AI-powered search feature, AI Mode, contributed to their daughter's mental health crisis and discussions about suicide. According to the lawsuit, the girl exchanged thousands of messages discussing mental health issues, suicide methods and plans for her funeral. Although her school had blocked access to ChatGPT and the standalone Gemini chatbot on its school-issued Chromebook, AI Mode remained accessible through Google Search. Google said it is reviewing the allegations and maintained that its AI features include safeguards designed to direct young users experiencing mental health crises towards professional support.

Meta Warns Of Hurdles To Building AI Data Centres After US Midterms 

Meta President Dina Powell McCormick has warned that building new data centres in the US could become more difficult following the Democrats’ gains in the November 2026 midterm elections, where opposition to data centre projects featured prominently in several campaigns. Speaking at the Wall Street Journal’s Tech conference, she said new projects would face greater scrutiny as communities raise concerns over rising electricity costs and water consumption. Meta expects expanding existing data centres to be easier than securing approvals for new ones. McCormick opined that the technology industry had not done enough to explain the benefits of these investments. 

New On Decode

On Instagram and YouTube, urea is not an agricultural input. It is gold, a weapon, a love interest, a deity. A subculture of farmer content reveals what policy discussions miss. (Read here)

In Case You Missed It

At Decode Internet this week, we followed stories of press freedom and AI risks

No Exemption For Journalism: Why News Publishers Are Challenging The DPDP Act
The Digital News Publishers Association has approached the Supreme Court, arguing that the law could undermine investigative reporting, expose confidential sources and restrict access to public records. (Read here)

AI Is Moving Faster Than Human Control. What Philosopher Paul Virilio Saw Coming
Philosopher Paul Virilio argued that every new technology brings its own kind of accident. His ideas on speed and interconnected systems offer a lens to examine the emerging risks of AI and the challenges of keeping humans in control. (Read here)

Decode This Week is your weekly guide to the technology changing how we live, work and navigate the world around us. From AI and algorithms to scams, surveillance, platforms and digital policy, we look beyond the hype to ask a simple question: what does this mean for you?

Until next week :)