A live federal procurement notice shows the Securities and Exchange Commission moving to renew a subscription to Babel Street's analytics platform, and the requirements attached to it go well beyond a database licence. The salient-characteristics document specifies "Agentic AI-Driven Workflows" - AI agents that run multi-step investigations, perform supply-chain vulnerability discovery, and analyze digital telemetry - layered on top of commercially sourced location, identity and public-records data. Offers are due August 5, with a proposed start date of August 14.
Key facts
- The record is a combined synopsis/solicitation on SAM.gov for a Babel Street Analytics software subscription renewal.
- Specified capabilities include Locate X Premium, the Insights platform, a US credit-header and criminal-record search, and identity-API calls.
- The requirement document explicitly calls for agentic AI workflows covering multi-step investigation, supply-chain vulnerability discovery and digital telemetry.
- The documents were first surfaced publicly by researcher Jack Poulson in his document-led report.
Why this is a different kind of story than "regulator buys software"
Three layers are being bought together, and the combination is the news. The first is commercially sourced data - location, identity, and public-records material acquired on the open market rather than compelled through legal process. The second is an OSINT platform that links and searches across it. The third, and the new one, is agentic automation designed to run investigative sequences without a human driving each query.
That third layer is what our explainer on AI agents describes in general terms, applied to a specific and consequential domain. An agent here is not a chatbot answering questions about a person. It is a loop with tools: pull the identity record, cross-reference the location history, expand the network, pull the telemetry, summarize. The value proposition for an investigator is throughput. The concern for everyone else is that throughput is exactly what used to be the practical limit on how many people a regulator could look at closely.
Babel Street itself markets its agents as tools for discovery, synthesis and analysis over what it calls mission-grade data. That is vendor positioning, and it should be read as such - it is not independent evidence of effectiveness, and it says nothing about how the SEC would actually use it.
What the record does not show
The distinction that most coverage will blur is the one worth holding: a regulator procuring an agentic commercial-data investigation stack is documented. A regulator already using AI agents to monitor Americans is not. This is a solicitation for a renewal, which documents intent to acquire, not a completed new contract or an operational system. Nothing in the public record establishes current use, scale, targets, or outcomes, and the SEC contact had not responded when Poulson published.
One term also deserves precision, because it gets stretched in retelling. A "credit header" is the identifying block attached to a credit file - names, addresses, phone numbers, dates of birth - as the Federal Trade Commission has described it. It is not a person's credit history. Reporting this as "the SEC is buying credit reports" would be wrong. Reporting it as "the SEC is buying a searchable identity graph" is closer.
The wider pattern
This lands in the same week as two other surveillance stories with an AI component. Delhi Police were observed running live camera feeds through facial recognition at a student protest site at Jantar Mantar; the Times of India reported four vehicles fitted with 360-degree cameras scanning crowds in real time against a database of more than 200,000 suspected criminals, while police said the purpose was identifying wanted criminals rather than monitoring protesters. And in the United States, organized opposition to Flock's automated license-plate reader network now visibly spans both conservative limited-government arguments and progressive civil-liberties arguments - Flock's own FAQ confirms its system indexes not only plates but vehicle make, model, color, timestamps, and visible features such as damage or aftermarket wheels, all searchable from a dashboard.
The common thread is not that AI invented any of this. Plate readers, face databases and data brokers all predate the current wave. What changes is the cost of asking a question. When a search that used to take an analyst a week becomes an agent task that runs unattended, the constraint that quietly limited surveillance - human attention - stops applying.
The honest caveat
There is real risk of overreading a procurement document. Requirements language often describes a vendor's full product catalogue rather than what a specific customer intends to switch on, and "brand-name-or-equal" renewals are routine administrative traffic. It is also entirely possible the SEC uses this stack narrowly, for the securities-fraud and market-manipulation work it is chartered to do. The document establishes what is being bought. It does not establish what will be done with it, and no one should pretend otherwise until the agency says.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)