Skip to main content

Risk Classifications

How Hermes Digital Screens and Flags Reputational Risk

Warning: Some of the descriptions below contain examples of language, imagery, or themes that may be offensive, graphic, or triggering. Content is reviewed only with informed consent.

What are social media screening risk categories?

Social media screening risk categories are the behavioural classifications used to flag reputational risk in a person's public digital footprint. Hermes Digital assesses content across 13 core categories — including prejudice, threats, drug imagery, nudity, profanity, self-harm, and political volatility — plus custom client keywords, with machine-learning probability scoring and human analyst verification.

Overview

Hermes Digital UK uses advanced machine learning and image recognition technologies to assess reputational risk in an individual's public digital footprint.

When a screening is commissioned, we retrieve content from the subject's associated social media accounts and evaluate it across multiple behavioural and contextual dimensions. This includes text posts, comments, likes, reposts, and images—each analysed against 13 core risk classifications and any bespoke keyword criteria provided by the client.

A post is flagged when any classification surpasses a predetermined probability threshold. For example, if a post shows a 65% likelihood of Toxic Language and a 73% likelihood of Hate Speech, it will be flagged under Hate Speech.

In cases where both a post and its associated image independently trigger risk criteria, both will be assessed and flagged accordingly. Reposts and likes are analysed by reviewing the original content in conjunction with any comments made by the subject. Replies and quote tweets are assessed based solely on the subject's own input.

Behavioural Classifications

The following categories define the types of digital content we flag during a screening:

1

Disparaging

Derogatory or demeaning statements targeting individuals or groups, often focused on appearance, intelligence, or character.

2

Drug Image

Photographic evidence of illegal drug use or paraphernalia, including depictions of pills, syringes, cannabis, smoking, or alcohol in illicit contexts.

3

Drug/Alcohol Mention

Written content referencing recreational drug or alcohol use, including coded slang and euphemisms.

4

Gory/Violent Image

Graphic visuals featuring blood, injury, disfigurement, crime scenes, or violence.

5

Nudity Image

Content involving explicit, pornographic, or suggestive nudity—including partial exposure that may breach broadcasting or employer standards.

6

Politics/Government

Strongly worded political opinions, commentary on government policy, or positions on contentious topics such as immigration, climate policy, or abortion.

7

Prejudice

Content containing racial, religious, homophobic, transphobic, or other discriminatory language or targeting.

8

Profanity

Use of vulgar, offensive, or obscene language across any form of communication.

9

Rude Gestures or Symbols

Visual gestures such as the middle finger, or images associated with extremist groups, hate symbols, or violent ideologies.

10

Self-Harm

Posts that reference suicidal thoughts, self-injury, or behaviours suggesting a risk to the subject's wellbeing.

11

Suggestive

Content with overt or implied sexual tone, sexual harassment, or posts that could be interpreted as sexually inappropriate in a professional context.

12

Threats

Statements implying violence, harm, or coercion—whether literal, implied, or metaphorical.

13

Weapons Image

Images of firearms, explosives, knives, or other weapons, particularly when shown in threatening or glorifying contexts.

14

Custom Keywords

Client-defined terms (positive, negative, or neutral) used to flag content containing specific language, themes, names, or organisations of interest.

Image Content Analysis

Our platform extends beyond text: it also evaluates images for embedded risks, including:

  • Violence, nudity, drug use, and extremist symbolism
  • Meme and poster text extraction using OCR (Optical Character Recognition)
  • Scene object recognition, e.g. identifying visual elements like syringes, alcohol bottles, protest signs, or branded items

For example, an uploaded image labelled with "Bicycle", "Uber", or "Electric", and text reading "Jump" may be flagged if any of those terms match a custom keyword profile supplied by the client. Hermes Digital UK supports client-defined terms, allowing screening to adapt to your regulatory environment or reputational thresholds.

Note: Image analysis is resolution-dependent. High-resolution files provide significantly more reliable and nuanced results.

A Note on Ethics and Interpretation

All findings are reviewed by qualified analysts to ensure contextual accuracy and prevent false positives. Hermes Digital UK adheres to a strict ethical code, and all screenings are performed with consent and data protection compliance under UK GDPR.

Risk Classifications — Frequently Asked Questions

What are social media screening risk categories?

Risk categories are the behavioural classifications Hermes Digital uses to flag reputational risk in an individual's public digital footprint. Our screening assesses content across 13 core categories — including prejudice, threats, drug imagery, nudity, profanity, self-harm, and political volatility — plus any bespoke keyword criteria supplied by the client. Each flagged item includes context, severity, and a probability score.

How are hate speech and prejudice detected?

Machine-learning models assign a probability score to each post for every applicable category. A post is flagged when a category surpasses a predetermined probability threshold — for example, a post scoring 73% for Hate Speech is flagged under Hate Speech even if it also scores for Toxic Language. Every flagged item is then reviewed by a qualified analyst to confirm contextual accuracy and remove false positives.

Does the screening analyse images and memes?

Yes. Beyond text, our platform evaluates images for embedded risk: violence, nudity, drug use, and extremist symbolism. OCR extracts text from memes and posters, and scene object recognition identifies visual elements like syringes, alcohol bottles, or protest signs. Image analysis is resolution-dependent — high-resolution files produce more reliable results.

Which languages does the screening support?

Text analysis covers 230+ languages, including coded slang and euphemisms. This lets us screen international executives, political candidates, and NGO personnel whose digital footprint spans multiple languages and regions.

What is a custom keyword profile?

A custom keyword profile is a set of client-defined terms — positive, negative, or neutral — used to flag content containing specific language, themes, names, or organisations of interest. It adapts the screening to your regulatory environment, sector standards, or reputational thresholds.

How are flagged items reviewed to avoid false positives?

All findings are reviewed by qualified analysts to ensure contextual accuracy and prevent false positives. Hermes Digital adheres to a strict ethical code: screenings are performed only with consent and under UK GDPR data protection compliance, and reports distinguish genuine risk from ambiguous or out-of-context content.

Commission a Screening

See these 13 risk categories applied to a subject's live digital footprint. AI-powered screening reports from £29.99 + VAT, delivered within 48 hours.

Explore Digital & Social Media Screening

Last updated June 2026