Sensitive coverage
Which clusters HermesAI flags as sensitive, what happens when it does, and why review can't be bypassed.
Some stories are higher-stakes than others. HermesAI flags those at the cluster level and routes them through editorial review — by design.
What fires the sensitive flag
The flag is set when a cluster involves any of:
Deaths or casualties
Specific people or counted incidents. Includes homicides, accidents, public-figure deaths, mass casualty events.
Political figures
Named politicians, government actions, election claims, partisan accusations.
Legal proceedings
Charges, court filings, sentences, trial coverage, ongoing investigations.
Medical or health claims
Drug efficacy, public health guidance, individual diagnoses, medical research claims.
Armed conflict
War coverage, military operations, displacement, casualties of conflict.
The flag is set at the cluster level — once the cluster is identified, every draft generated from it inherits the flag.
What happens when the flag fires
Visible in the feed
A clear sensitive marker on the story card. You see it before opening.
Visible in the story workspace
The marker appears prominently. Compose and refine work normally; the AI knows the context.
Cannot self-publish
Even reporters with self-publish permission for non-sensitive coverage cannot self-publish a sensitive draft. It must route through editorial review.
Routes to editorial review
On submit, the draft enters in_review. A reviewer has to approve before it can be delivered.
Review surface shows the reasoning
The review workspace shows why the cluster was flagged sensitive — which trigger fired, which sources used the trigger language.
Why review can't be bypassed
No override
There is no admin toggle to disable sensitive-topic review. There is no plan tier that removes it. There is no per-story override.
This is intentional. A newsroom that can disable editorial review on sensitive coverage is a newsroom that will, eventually, do exactly that under deadline pressure. HermesAI removes that option.
What sensitive flag is not
Not a quality signal
Sensitive doesn't mean low confidence. A high-confidence cluster on a sensitive topic is still routed to review.
Not a slowdown for everything
Non-sensitive clusters publish normally. Only stories involving the listed topics route through review.
Not a content filter
Sensitive doesn't mean 'don't cover this'. It means 'an editor signs off'.
Not topic blacklist
HermesAI covers sensitive topics. It just makes sure a human is in the loop on each draft.
How the flag is decided
A combination of signals from the cluster:
- explicit mentions in source coverage (death tolls, charges filed, named politicians, etc.),
- topical category match (legal, conflict, health),
- pattern matches in source content (medical claims, electoral claims).
If you see a sensitive flag on a cluster that doesn't seem to involve any sensitive content, that's worth flagging — surface it via the Contact channel. Mistakes happen; the catalog evolves.
What a reviewer should do
Read the trigger
The review workspace shows which trigger fired. Start there — is the trigger right?
Read provenance harder than usual
Sensitive stories deserve more attention to source quality. Tier 1 + wire coverage gives you firmer ground.
Watch for conflict markers
Sensitive + conflicting sources = the editor's call really matters. Decide deliberately.
Approve, reject, or escalate
Standard outcomes. When in doubt, escalate to the editor-in-chief.