Large language models for influence and market manipulation

Large language models (LLMs) for influence and market manipulation are generative text systems that can produce varied content for narrative operations, impersonation, sentiment campaigns or false market-relevant claims. They can reduce the labour needed to draft and adapt text, but generation is separate from distribution by bot networks, audience response, trading activity and any legally established manipulation.

Function

An LLM can draft posts, comments, articles, analyst-style notes or messages adapted to an audience, language or platform. Variation can make simple duplicate-text detection less effective, although provider records, account behaviour and distribution patterns may still support investigation. In finance, possible uses include coordinated sentiment, fabricated research and impersonation of officials or institutions. Capability does not establish use, reach or effect.

Strategic significance

OpenAI's 30 May 2024 threat report described the disruption of five covert influence operations that used its models for tasks including content generation. The provider assessed that these operations did not achieve meaningful audience engagement or reach through its services. This is evidence of model use in influence activity, not evidence of a financial campaign. Similar text could be directed at investors or paired with automated trading, but a manipulation finding would require evidence of actor, intent, dissemination, trading conduct and market effect.

Control or weaponisation history

Countermeasures include usage-policy enforcement, account and network takedowns, provenance tools, disclosure by model providers and verification inside financial institutions. Their effectiveness varies, particularly where models are open-weight or hosted outside a platform's jurisdiction. In Economic Kill Chain terms LLMs can assist positioning and amplification by producing variants of a narrative at low cost. That is an enabling capability, not evidence that a state campaign exists or that generated content changed a market outcome.

Current status and evidentiary limits

ESMA's analysis of 20 February 2026 documented AI adoption and market trends, while FINRA's notice of 18 June 2026 and the CFTC advisory described fraud and communications risks. These materials do not establish that a model output caused a price move or that any actor committed manipulation. Analysis must separate model capability, generated content, distribution, audience response, trading activity and adjudicated intent. The SEC's 2024 enforcement actions against misleading claims about firms' use of AI illustrate a different offence, disclosure to investors, rather than proof of an LLM-directed market campaign. ESMA reported that surveyed securities-market use was concentrated in lower-autonomy applications, including internal and back-office functions. Adoption evidence therefore should not be converted into a claim that autonomous influence or trading systems are widespread.

See also

Coordinated social-media bot and amplification networks · Deepfakes and synthetic media in financial disinformation · Disinformation and market manipulation · Reflexive control in financial markets · AI-enabled market manipulation · Amplification (EKC Phase 5) · Economic Kill Chain (EKC) · Economic statecraft · Positioning (EKC Phase 3)

Sources

• CFTC customer advisory on AI fraud.
• FINRA, deepfakes and vishing, 18 June 2026.
• SEC enforcement release 2024-36.
• ESMA, AI adoption and trends in securities markets, 20 February 2026.
• OpenAI, AI and Covert Influence Operations, 30 May 2024.