What is adverse media screening?

Adverse media screening is the check of news, public records and other open sources for negative information about a customer or counterparty, such as links to fraud, corruption, organised crime or sanctions evasion. It feeds a firm's view of customer risk. It deals with unverified text, so most of the work is judgement.

Why does adverse media screening matter once agents act?

An agent can read far more sources than an analyst, in more languages, and faster. That is the attraction. The risk is that it treats everything it reads as fact. A news article is external information: someone else's claim, published for their own reasons, often about a different person with the same name.

So the agent's job is to find, sort and summarise, not to conclude. This is the second-line work compliance agents are built for, and whether an AI agent can do adverse media screening depends on keeping the conclusion, and anything that changes the customer's treatment, with a person.

How does it work in practice?

Screening runs at onboarding, at periodic review and when something triggers a fresh look. A typical pass has four steps:

1. Search sources for the customer's name, aliases and associated entities.
2. Discard hits that are plainly about someone else, using date of birth, location, role and other identifiers.
3. Classify what remains by severity and credibility: a court judgment carries more weight than an anonymous blog.
4. Summarise each relevant hit with its source, date and the specific allegation, for an analyst to assess.

Our guide on how to run adverse media screening with an AI agent sets out who acts at each step.

An example from onboarding

Take a business bank onboarding a logistics company. The agent finds twelve articles mentioning a director's name. Eight concern a footballer with the same name and are discarded with the reason recorded. Three are trade press about the company's growth. One is a regional news report that the director was questioned in a customs fraud inquiry several years ago, with no charge reported.

The agent drafts a note quoting the allegation, citing the source and marking it unverified. It does not rate the customer. An analyst reads the article, checks for any later outcome and decides whether enhanced checks are needed.

How is adverse media screening different from sanctions screening?

Sanctions screening matches names against official lists published by governments. A confirmed match has legal consequences: you generally cannot deal with that person. The source is authoritative, the question is narrow, and the analyst's task is mostly to confirm or rule out identity.

Adverse media has no list and no binding consequence. A negative article does not prohibit a relationship; it informs a risk judgement. The sources range from reliable to worthless, and the analyst must weigh credibility as well as identity. That is why agents that read sanctions screening matches and agents that read the news need different briefs and different floors.

Which terms sit next to adverse media?

• PEP screening identifies politically exposed persons, who often appear in adverse media for reasons unrelated to wrongdoing.
• False positive: a hit about someone else. Adverse media produces many, and reducing false positives with an AI agent is where agents earn their keep.
• Negative news: another name for the same check.

Where Pryme fits

In Pryme, every article an agent reads arrives tagged as external, the weakest provenance class. External material can inform a draft but cannot, on its own, support a decision about a customer. The agent's summary is held for a named analyst, and the filed record shows which sources were read, which were discarded and why.