AI isn’t a just trading tool; its behaviour is now shaping how markets function. Across asset classes, AI systems are influencing liquidity formation, volatility dynamics, and price discovery. Rapid sector repricing following major model releases, emerging signs of coordinated agent behaviour, and the migration of liquidity risk into funding and leverage channels all point to AI becoming embedded in the mechanics of secondary markets rather than sitting at the edge of trading technology. Here’s what I learnt this week on AI in Trading:

1. Claude and the acceleration of the “AI trade”

Anthropic’s release of a legal-automation plug-in for Claude Cowork AI triggered a rapid $285bn sell-off in European software stocks, as investors repriced firms exposed to legacy software and data workflows amid fears of accelerated AI disruption:
https://www.reuters.com/business/media-telecom/ai-concerns-pummel-european-software-stocks-2026-02-03

This was reinforced days later by the launch of Claude Opus 4.6, which introduced a one-million-token context window and coordinated multi-agent capabilities, intensifying concerns about the long-term viability of traditional SaaS models:
https://www.ft.com/content/a0cd0281-8367-4ed3-9f18-038e4a9f79e0

While sector rotations are not new, the speed, volatility, and correlation of this repricing were notable. AI-driven predictive systems now reprice volatility surfaces in real time, while liquidity-provision algorithms dynamically reallocate capital across equities, derivatives, and ETFs. Once volatility thresholds are breached, liquidity is often withdrawn conditionally rather than reactively, leading to sharp execution asymmetry as spreads widen and depth disappears.

Research shows AI and machine learning improve market efficiency under normal conditions by accelerating price discovery, narrowing spreads, and reducing transaction costs. Under stress, however, synchronised algorithmic behaviour and feedback loops can amplify short-term volatility – underscoring the need for stronger governance, a theme discussed in depth at this week’s FIX AI Working Group webinar (see the end of the newsletter).

2. Moltbook and coordination risk in agent-based systems

Last week’s focus on Moltbook – the social network where agents appeared to debate consciousness and coordinate behaviour – shifted this week. Reports suggest some of the most visible activity was human-prompted rather than driven by genuine agent-to-agent interaction:
https://www.moltbook.com/
https://x.com/HumanHarlan/status/2017424289633603850

Growing interest in Moltbook is not agent sentience, but what it reveals about coordination effects when many agents operate within a shared environment. This is directly applicable to secondary-market trading, where firms operate across fragmented venues, asset classes, and regulatory regimes. When AI is deployed as tightly coupled, model-centric point solutions connected directly to execution, experimentation feeds straight into markets. This increases operational risk, correlated behaviour, duplicated security logic, fragmented audit trails, and high coordination costs when models or prompts change – failure modes commonly seen in real-world AI deployments. In contrast, the argument is that a gateway-based architecture provides an alternative by routing AI activity through a shared control layer, enabling consistent observability, unified risk and capital controls, and the ability to scale only once behaviour is understood. Read more here – https://lnkd.in/gMjRezyn

3. Liquidity risk is shifting into funding and leverage

Alongside the AI-driven equity sell-off, the Financial Stability Board published a report highlighting that the $16tn government-bond-backed repo market is now dominated by funding, leverage, and collateral dynamics rather than execution venues:
https://www.fsb.org/2026/02/fsb-warns-of-financial-stability-challenges-in-repo-markets/

The FSB points to structural vulnerabilities including high leverage in non-centrally cleared segments, extensive collateral reuse, concentration among key intermediaries, and pro-cyclical margining. When repo funding tightens via higher haircuts, margin calls, or lender withdrawal – leveraged participants can be forced into rapid asset sales, transmitting stress directly into secondary-market prices and impairing dealer intermediation. Episodes such as the 2019 US repo spike and the 2022 UK gilt crisis show how quickly funding shocks can overwhelm even highly liquid markets. As electronic liquidity providers and AI-driven risk systems become more dominant, these effects are amplified. Liquidity now evaporates non-linearly across markets, making resilience a balance-sheet and funding problem – not just an order-book or technology issue.

4. Implications for trading technology

Recent developments around Claude also highlight a broader shift: AI is collapsing the cost, time, and skill barriers to building bespoke trading tools. The historic platforms are increasingly being challenged by personalised, modular stacks that replicate only what is needed, weakening vendor lock-in and compressing incumbent margins. Trading systems are evolving from static information displays into adaptive workflows that learn a trader’s style through trade logs and feedback loops, improving discipline and decision quality. By scaling pattern recognition and market scanning across thousands of assets, future use of AI in trading is more likely to support AI-augmented discretion than fully autonomous agents. As data access becomes commoditised, competitive advantage shifts to interpretation and synthesis – the trader’s edge:
https://www.linkedin.com/posts/haroldberens_the-digital-gutenberg-moment-activity-7425544430163689472-b978

5. First FIX AI Working Group webinar

This week also saw the first FIX AI Working Group webinar, organised by FIX APAC, with @GaryAng (ex MAS), @PetrosKyliakoudis (Baillie Gifford), @NickIdelson (TraderServe / FIX Algo Working Group), and @PJDiGiammarino (Reg Risk Solutions).

The discussion focused on how AI – particularly agentic systems – is moving from experimentation into live trading workflows. While fully autonomous execution remains limited, the growing use of AI for recommendations, strategy selection, workflow orchestration, and decision support increasingly raises critical questions around control, accountability, and market integrity.

The panel emphasised that rather than new risks – AI amplifies existing ones: non-determinism, feedback loops, data quality issues, and third-party dependencies. Even traditional deterministic algorithms can destabilise markets due to latency and ecosystem interactions; AI increases these risks and heightens the importance of robust governance, testing, and standards.

From a regulatory perspective, AI needs to be treated as a material risk capability, not a standalone experiment. Supervisors are less focused on the specific model used and more on whether firms can demonstrate control, explain outcomes, and remain accountable. This includes maintaining a clear inventory of AI use cases, understanding levels of autonomy, and applying proportional controls as systems become more independent.

Data quality and standardisation – such as shared semantics via FIX standards – emerged as foundational. Firms must be able to reconstruct events end-to-end, capturing inputs, model versions, recommendations, human decisions or overrides, timestamps, and execution actions. Properly constrained agentic systems may ultimately be more auditable than unconstrained generative AI. Full Webinar recording and full write-up to follow.

Key takeaways

The central lesson from this week is not simply that AI adoption in trading is accelerating, but that AI behaviour is becoming a market variable. Predictive models, agentic workflows, and AI-driven risk systems now influence when liquidity appears, when it disappears, and how stress propagates across markets.

For firms, the implication is clear: AI must be governed as a core risk capability, not an experimental overlay. Competitive advantage will come from disciplined deployment – clear accountability (including board oversight), rigorous testing, high-quality data, strong observability, third-party risk management, and adherence to industry standards such as FIX.

AI is not a shortcut around controls. How well firms govern AI behaviour will increasingly determine not only their own outcomes, but the stability and integrity of secondary markets themselves. The next phase of market-structure reform is likely to focus less on trading rules, and more on leverage, transparency, and the resilience of the financial plumbing on which AI-driven markets now depend.

Thanks for reading – as always, feedback always welcome.
Rebecca