In short: Adverse media review, also called negative news screening, is the work of checking whether press coverage and public records about a customer indicate financial crime risk. Most of what a screening tool surfaces is noise: a namesake, a story where your customer is the victim rather than the subject, or an article too old to bear on today's risk. Zyphe runs the review as an operated desk. Agents read the underlying sources, resolve whether the coverage is genuinely about your customer, classify it against your risk taxonomy, and produce a cited summary. Your analysts decide what it means for the relationship.
Desk description last reviewed 2026-08-20.
What arrives at this desk?
The work this desk takes off your team, stated as the queue it actually receives.
Onboarding negative news hits
Coverage surfaced when a new customer is screened, where the question is whether anything in it should change how you onboard them.
Periodic review sweeps
Scheduled re-screening across an existing book, where the volume is high and the proportion of genuinely relevant coverage is low.
Event-driven hits
A story breaking on a customer you already hold, where the timeline between publication and your response is the thing a supervisor will look at.
Unclear escalations
Cases where the coverage is real and the entity is right, but whether it is material to your risk appetite is a judgment call.
How does the adverse media review desk work?
Five steps, the same way every time, because consistency is what an audit is actually testing for.
- 1
Read the source, not the headline
Aggregated snippets routinely drop the qualifier that changes the meaning: charges dropped, a person named as a witness, an allegation attributed to a party rather than established. The agent opens the underlying item and works from its text.
- 2
Resolve the entity
The agent establishes whether the person or company in the coverage is actually your customer, using date of birth, location, employer, corporate identifiers and the surrounding detail, and records what the resolution rests on.
- 3
Classify against your taxonomy
Each relevant item is placed against your own categories: predicate offence type, the customer's role in the story, the date, and the credibility of the outlet. Your taxonomy, not a generic risk label.
- 4
Summarise with citations
The customer gets one summary that a reviewer can act on, with every assertion linked to the item it came from, so nothing rests on the agent's paraphrase alone.
- 5
Escalate the material findings
Anything that plausibly changes the risk picture goes to your analyst with the reading already done and the open question stated plainly.
What we prepare, and what you approve.
Agent-prepared, human-approved is not a hedge in the marketing copy. It is the shape EU law requires, and it is the reason this desk can be bought by a regulated firm at all.
| Stage | The agent prepares | Your team approves |
|---|---|---|
| Screening noise | Discounts namesakes and irrelevant coverage, with the reason recorded | Approves the discount, or pulls it back |
| Materiality | Classifies coverage against your taxonomy and explains the classification | Decides whether the finding is material to your risk appetite |
| Customer risk profile | Assembles the findings that bear on risk | Sets or revises the risk profile AMLR Art. 18(3)(c) |
| The relationship | Prepares the file for a decision | Decides whether to onboard, continue, restrict or exit AMLR Art. 18(3)(d) |
| Suspicious activity | Drafts the narrative from the coverage and the account history | Decides whether to report, and files with the FIU AMLR Art. 18(3)(e) |
Why the split is drawn there
Article 18 of Regulation (EU) 2024/1624, the EU anti-money laundering regulation that applies from 10 July 2027, permits obliged entities to outsource tasks to service providers, and requires the supervisor to be notified before the provider starts. Article 18(2) then treats the service provider as part of the obliged entity, leaves the entity fully liable, and requires it to be able to demonstrate to its supervisor that it understands the rationale behind the activities the provider carries out. Article 18(3) lists tasks that cannot be outsourced under any circumstances. Every one of them sits on your side of this table.
That demonstrability requirement is the reason each case leaves this desk with its reasoning written out rather than with a score. A confidence number cannot be explained to a supervisor. A rationale can.
What this desk will not do
- Define your adverse media taxonomy or lookback period
- Decide what is material to your risk appetite
- Exit or restrict a customer relationship
- File a suspicious activity report on your behalf
- Treat an unverified allegation as an established fact
What do you actually receive?
A decision-ready file, in the systems you already run.
- One cited summary per customer, not a list of links
- A recorded reason for every item discounted
- Entity resolution stated explicitly, including what it rests on
- A queue worked to an agreed service level, including event-driven hits
The software behind this desk
If you would rather run the work yourself than have it operated, these are the same capabilities as a product surface.
Frequently asked questions about adverse media review.
What is adverse media screening?
Adverse media screening, also called negative news screening, is the practice of searching press coverage and public records for information suggesting a customer is involved in financial crime or related wrongdoing. It runs at onboarding, on a schedule during the relationship, and on trigger events. The screening itself is mechanical. The work is in the review: deciding whether the coverage is about your customer at all, and whether it is material.
Why do adverse media tools produce so many false positives?
Because they match on names against an enormous corpus with very little context. A common name in a populous market can return thousands of items, almost none of which concern your customer. Tools also surface coverage where the customer appears as a victim, a witness, or an unrelated mention, and they rarely distinguish an allegation from a conviction. None of that is a defect the vendor can fix by tuning: it is the point at which a human reading is required.
Is adverse media screening a legal requirement in the EU?
It is not a standalone statutory obligation with its own article. It is one of the ways firms meet due diligence and ongoing monitoring duties under the EU framework, and supervisors expect it where the risk profile warrants it, particularly for enhanced due diligence and for politically exposed persons. Your own policy is what defines the scope, the sources and the lookback period, and that policy stays yours.
How far back should adverse media coverage be checked?
There is no fixed period in the EU framework, which is why the lookback belongs in your policy rather than in a vendor default. What matters for an audit is that the period is stated, justified against your risk assessment, and applied consistently. The desk applies the period you set and records where an item fell relative to it.
Do you use an LLM to summarise the articles?
Agents read and classify the source items, and the summary is generated. That is precisely why every assertion in it is linked back to the item it came from, and why the summary is prepared for a reviewer rather than treated as the decision. Under AMLR Article 18(2) you have to be able to demonstrate that you understand the rationale behind what your provider does, and a citation you can open is the only version of that which survives a supervisor asking.
The other desks
Most teams start with one queue and add the next once the first is clearing.
Sanctions and PEP alert review
Works the screening queue your engine produces: compares each hit against your matching policy, writes the disposition and its reasoning, and escalates the ones that are genuinely unclear.
See the deskUBO and EDD review
Walks the ownership chain through corporate registries to the natural persons behind it, screens them, and assembles the enhanced due diligence file including what could not be resolved.
See the deskTransaction monitoring alert triage
Reconstructs the account history behind each rule-triggered alert, tests it against that customer's own baseline and your typologies, and drafts either a reasoned close or a narrative your investigator can amend and file.
See the deskKYC periodic review and refresh
Re-runs the checks on customers falling due, compares what it finds against the file you already hold, and reports what actually changed rather than rebuilding a dossier nobody reads.
See the deskBook a demo
Put the adverse media review queue on a desk.
Book a demo and bring a real slice of your queue. We will work it, show you the files, and you will see exactly what your team still has to approve.