What happens when artificial intelligence can execute stock trades as easily as it browses the web? A significant regulatory blind spot is emerging. As AI-powered browsers and desktop automation tools become mainstream, they’re enabling a new form of trading that sidesteps virtually every safeguard regulators have built around algorithmic trading. This isn’t a distant possibility—it’s happening now, and it threatens to upend years of carefully constructed market oversight.
With the rise of AI browsers such as Perplexity’s Comet and OpenAI’s ChatGPT Atlas, and tools like BrowserUse or agentic AI tools which can control and operate the browser or desktop at large, an interesting use case has emerged: AI trading. Instead of the API or algo route, users can simply log in to their terminals and write strategies in simple, prompt-style syntax, after which the Large Language Model (LLM) takes over the constant monitoring or trading on behalf of the user.
Fundamentally, High Frequency Trading (HFT) or algo trading at large consists of a research component, wherein quantitative researchers attempt to write strategies based on past market momentum to predict price action. This also includes simpler strategies such as arbitrage between securities on different exchanges, or attempts by the algo to price derivatives most accurately relative to their underlying asset class using perhaps Black-Scholes or Binomial models, depending on whether it’s a European option or an American option. Further, having the fastest access to market data enables these firms to make a few cents on the dollar for every trade, which accumulates into an economies-of-scale model when done on a constant basis. At the heart of these strategies lies a common factor: the fastest execution and most efficient order placement in order to time the market. In India, approximately 55% of total trading volume is algo-driven, compared to 75-80% in the United States.
SEBI’s Constant Attempt to Regulate Algorithms
SEBI’s regulation of algorithmic trading traces back to its 2008 circular on Direct Market Access (DMA), which introduced technology-driven order placement by institutional investors and set the stage for formal oversight of automated strategies in India. Early measures culminated in SEBI’s 2012 and 2013 “Broad Guidelines on Algorithmic Trading.” As co-location and high-speed access expanded, SEBI’s 2018 measures further strengthened the framework, addressing order-to-trade imbalances, throughput controls, and proximity hosting. Finally, in 2025, SEBI pivoted to a dedicated retail framework that centers on broker accountability, exchange approval and unique tagging of algos, static-IP secured API access, and categorization of white-box and black-box strategies, while phasing implementation timelines to support safer participation.
However, it seems the nearer we race to the implementation deadline, we are simultaneously racing toward the obsolescence of the same, by hitting an inflection point wherein AI trading would be easier than algo trading (for retail clients).
How AI Trading Bypasses All Regulatory Checks on Algos
There are specific restrictions placed on the flow of algo orders, and a few corresponding to normal orders. AI trading essentially acts as a wolf in sheep’s clothing, since the trading front would be the user terminal itself, thereby bypassing any checkpoint intended for automated orders.
An order that originates through a registered algorithm is mandated to be tagged as an algo order. This ensures demarcation and monitoring of both the algos and the orders fired. With AI trading, the executing program would not be within the purview of exchange monitoring, and orders would actually be tagged as normal orders.
Further, algos are restricted to Order Per Second (OPS) thresholds, whereas there is no such restriction on normal orders, and therefore no corresponding restriction on AI trading. Additionally, algo orders have an enhanced ability to manipulate order books and markets by creating noise—i.e., placing high fictitious orders otherwise known as “Persistent Noise Creators.” When orders are tagged with an algo ID, exchanges have the ability to keep a check on order modifications and cancellations within permissible thresholds. Further, there are checks for synchronized trading and structured purchasing, which can be easily bypassed using AI trading.
To Permit AI Trading or Keep the Route Restricted to Algos?
Retail investors are far removed from executable algos. Even the present do-it-yourself algo platforms are essentially strategy builders in the cloud, and not research-oriented algorithms that can factor in myriad market conditions. AI democratizes this by eliminating the need for knowing obscure languages like OCaml or even easier ones like Python, and enables users to set up multiple data points that can be constantly fetched and analyzed. In light of this democratization, a fundamental debate arises: if we’re concerned about AI bypassing regulatory checkpoints and creating information and execution asymmetry, shouldn’t we be equally concerned about co-location and DMA? After all, institutional players have long enjoyed similar advantages—just packaged in a more sophisticated, expensive manner that’s been deemed acceptable.
However, an outright ban on AI trading would be both impractical and counterproductive. The technology is already here, and attempting to prohibit it would simply push the activity underground or offshore, beyond regulatory reach entirely. The more pragmatic path forward lies in leveling the playing field through smart, technology-neutral regulations.
The simplest solution is to shift from investor-type-oriented regulations to universal technical safeguards. For instance, an Order Per Second (OPS) threshold could be set for all investors—whether using AI, registered algos, or manual trading. Orders exceeding this threshold could be delayed by one second or so, creating a uniform speed limit that prevents market manipulation regardless of the technology employed. This approach would eliminate the regulatory arbitrage that currently makes AI trading attractive precisely because it’s unregulated.
Further, from a regulatory standpoint, it’s necessary to consolidate algo and AI trading laws into a unified framework. The current bifurcation—where registered algos face strict oversight while AI-driven trading faces none—is untenable. What’s needed is a baseline minimum allowable criteria that applies universally, without burdensome registrations or restrictions that stifle innovation. This would allow retail investors to freely experiment and participate, while still maintaining essential market safeguards against manipulation and systemic risk.
The stakes are clear: without action, we risk creating a two-tiered market where sophisticated AI users can exploit regulatory gaps that their algo-trading counterparts cannot. The window to address this is narrow—once AI trading becomes widespread, retrofitting regulations will be far more disruptive than getting ahead of the curve now.
