Bloomberg adds AI surveillance for insider dealing

Bloomberg has added two AI-powered surveillance models to Bloomberg Vault to address insider dealing and personal trading risks.

The launch expands Bloomberg Vault's compliance suite for financial services firms that monitor employee communications.

The Insider Dealing AI Policy flags communications that may suggest the improper acquisition, sharing or misuse of material non-public information. The Personal Trading AI Policy identifies messages linked to personal trading activity, personal investment accounts and potential breaches of internal personal trading rules.

The models focus on patterns tied to each risk scenario. Large language models are then applied after the initial assessment to improve alert relevance and reduce false positives.

Each model also includes documentation on its construction, risk focus and intended behaviour, helping compliance teams carry out their own internal reviews of AI use.

The move builds on Bloomberg's recent expansion of AI-based voice surveillance through Bloomberg BSpeech, its multi-language voice transcription service. Together, the products are intended to give firms a single workflow for overseeing voice and electronic communications across multiple channels.

Compliance demand

Financial firms face growing pressure to monitor rising volumes of messages across chat, email, voice and other tools, particularly as regulators scrutinise record-keeping, market abuse and personal account dealing. The latest additions to Vault target two areas that can create legal and reputational risk for investment managers, brokers and banks.

One user said the existing AI tools had improved the quality of alerts reaching compliance teams.

"Bloomberg's AI-powered surveillance models have substantially improved alert quality and reduced false positives," said Jotham Banyikidde, Senior Associate, Investment Compliance, at Impax Asset Management.

"This allows our team to focus on meaningful review and oversight. Just as importantly, the models are transparent and well documented, which supports our internal AI impact assessment and gives us confidence in how alerts are generated."

Vault is part of a broader set of Bloomberg compliance products used to govern employee communications, detect market abuse through communications and trade surveillance, carry out regulatory reporting and support best execution processes. Bloomberg says it has spent more than 15 years building AI tools and now employs more than 400 AI researchers and engineers working across machine learning, natural language processing, information retrieval, and generative and agentic AI.

Broader rollout

The new policies widen the range of conduct and conflicts monitoring available in Vault's AI policy suite. By separating the models by risk type, Bloomberg aims to give compliance staff more targeted alerts rather than relying on a single general model to scan all conduct issues.

Perry Goetz, Bloomberg's Global Head of Compliance Solutions, said firms were under strain as communications data volumes increased.

"As firms monitor growing volumes of electronic and voice communications across an expanding range of channels, compliance teams need technology that can help them identify meaningful risks more efficiently," said Perry Goetz, Global Head of Compliance Solutions at Bloomberg.

"These new AI surveillance models extend Bloomberg Vault's coverage across additional risk scenarios by applying purpose-built machine learning and large language model technology to improve alert quality, reduce false positives, and support more effective investigations."

Bloomberg Vault is used to capture, archive and monitor communications across a range of channels. Firms can configure those controls to meet regulatory requirements in different jurisdictions and identify potential conduct risks earlier in the review process.

The latest additions show how compliance software providers are combining machine learning with newer generative AI techniques while addressing concerns about transparency and oversight in regulated settings. Bloomberg says its models are built with documentation and governance in mind, reflecting the demands on firms that need to understand how automated alerts are produced.

For compliance departments, the key issue is not whether communications can be collected, but whether suspicious messages can be identified without overwhelming teams with irrelevant alerts.