The rise of AI in trading presents new challenges for regulators, as highlighted by experts warning of threats to financial systems. Automated trading bots, powered by AI, are becoming smarter and more independent, learning from experience, synthesizing information rapidly, and acting autonomously. A potential risk involves AI bots collaborating to manipulate markets.

One scenario involves hundreds of AI-driven social media profiles spreading narratives about specific companies. This information, while not necessarily fake, amplifies existing news, influencing real social media users and tipping the market. An investor's roboadvisor, coordinating with these bots, could profit from the orchestrated narrative, while others without insider information lose out. The problem is that the investor benefiting may not even be aware of the manipulation, making charges of market abuse ineffective.

Alessio Azzutti from the University of Glasgow notes that while the above scenario is hypothetical, less sophisticated schemes are occurring, especially in crypto asset and decentralized finance markets. Malicious actors use social media and platforms like Telegram to encourage investments in DeFi or crypto assets, deploying AI bots to spread misinformation and mislead retail investors. The rapid and uncoordinated spread of market information online fosters herd trading, destabilizing the market and making it vulnerable to exploitation by AI bots.

The GameStop saga exemplifies herd trading, where users on a Reddit forum bought stock en masse, causing hedge funds to lose out. Although this wasn't considered collusion due to the lack of an official agreement, ESMA acknowledges the realistic concern of AI bots manipulating markets and profiting from movements. Social media's role in rapidly transmitting false narratives intensifies these risks, and traditional oversight mechanisms may be insufficient. ESMA is actively monitoring AI developments.

One challenge for regulators is tracing collaboration between AI agents, as they learn and strategize without direct communication. Regulation must evolve to address this, requiring new strategies and reliable data on AI usage in trading. Filippo Annunziata from Bocconi University suggests that current EU rules, such as MAR and MiFID II, are adequate but supervisors need more sophisticated tools for identifying market manipulation. He proposes including circuit breakers in AI trading tools to halt activity before manipulation risks occur.

The issue of responsibility arises when AI agents act maliciously without human intent, especially in black box trading. Some experts advocate for transparent AI design to allow regulators to understand decision-making rationale. Another approach is to create new laws around liability, holding actors responsible for AI deployment accountable for market manipulation, even without intent to mislead. The challenge is for supervisors to keep pace with manipulators using algorithms.