If you’ve been tapping TikTok‘s “Not Interested” button hoping to reclaim your feed, new research suggests the effort has a shorter shelf life than the platform implies. A study from Northeastern University’s Khoury College of Computer Sciences found that the feature provides only temporary relief, with unwanted content resurfacing within minutes after a user stops using it.

How the researchers tested it

The team, led by Professor Piotr Sapiezynski and doctoral student Levi Kaplan, ran a sock puppet audit on TikTok’s mobile app across three phases. They created automated accounts and trained each on cooking, fitness, or sports betting by having them watch over 200 videos on each topic. The accounts were then split into groups: one marked videos as “Not Interested,” and the other simply swiped past them.

In the short term, the button worked better than swiping in the majority of test runs across all three categories. But the long-term results told a different story. Once the accounts stopped pressing the button, the flagged content returned quickly, with cooking videos, the study’s baseline category, rebounding especially fast.

The bigger concern

The researchers say the core problem is consistency. The button’s effectiveness varied significantly by topic, and users had no reliable way to predict when or why it would stop working. Kaplan noted that while the button is still better than doing nothing, TikTok users who find their feeds still dominated by content they’ve actively rejected are experiencing a documented platform behavior, not a glitch. Sapiezynski said the findings could be relevant for regulators in the European Union, where offering a control feature that doesn’t perform as described may qualify as a dark pattern under the Digital Services Act.

The study adds to a growing body of research questioning whether social media platforms’ user-facing controls deliver what they promise. The researchers have called on TikTok to make the feature work consistently across all content categories.

As AI agent traffic surpasses human traffic on the internet, companies working in commerce and finance are building tools that allow agents to take action on behalf of users at a rapid pace. Days after trading platform Robinhood introduced agents that can trade for users, Coinbase launched its own agents that can execute trades and pay for premium research.

The company said Thursday that users can integrate the agent with their main account and start trading. If users don’t want to give the agent access to their main account, they can choose to have it operate in a separate sandbox.

Coinbase noted that the agent can use tools like Coinbase Advanced, the company’s platform for professional traders that includes extra features like TradingView charts to analyze and execute trades. Users can ask the agent to rebalance their portfolio, ask it to follow an investment thesis and trade on their behalf, or provide advice on a one-time crypto trade.

At the moment, the agent can trade in crypto spot markets and derivatives, with support for equities and prediction markets planned for the future. Coinbase added that it will soon add support for custom limits such as maximum trade size, which services the agent can interact with, and how much it can spend.

Coinbase is taking advantage of the open x402 payment protocol it launched in collaboration with AWS, Anthropic, Circle, and Near last year. Using this standard, the agent can pay for premium research data APIs and on-demand compute for trading insights without requiring any login or subscription. This website lists services that the agent can access through the x402 protocol.

The trading platform has been actively investing in AI tools for the last few years. It launched AgentKit, which allows developers to integrate automated wallets into their apps in 2024. Last December, it added an AI-powered assistant to the app that provides trading tips and financial advice. The company said that the latest agent launch can also work in ChatGPT or Claude through its MCP server.

“Coinbase for Agents is informed by insights gleaned from years of building the agentic economy, and the primary goal is to create agents that can transact. And unlike pure trading platforms, we’re the only one that combines exchange access with a native payments protocol. We’re aiming to build a fundamentally different product for a future where most of the internet is accessed through agents,” Lincoln Murr, Head of AI Product, told TechCrunch via email.

AI companies are exploring agentic payments at a rapid pace through new partnerships. Last month, Visa invested in Replit to power agentic payments for developers. The payment network company made a deal with OpenAI this week to explore similar products. The pace of development in the sector has made global financial regulators take notice. The Financial Stability Board (FSB) said that there should be strong safeguards in place to mitigate AI risks.