AI Empathy 1.4 and 1.5: Governance Policies, Agent Tools, ECA and a One-Step Recipe for Drupal
Two releases of the AI Empathy Evaluation module for Drupal are out. Version 1.4 adds an AI governance policy layer: choose the frameworks that apply to you and score every AI response for how well it honours them. Version 1.5 takes empathy checks out of the evaluation dashboard and puts them where AI already runs on a Drupal site: AI agents, the AI Assistant, ECA workflows, guardrails and text fields, with a recipe that sets it all up in one command.
This post walks through both releases, and shows how the module fits with the rest of the Drupal AI ecosystem.
1.4: AI governance policies
In August I argued that your AI governance policy is a PDF that your prompt has never read. Version 1.4 is the answer to that problem, built as six pieces of work under work item #3608922.
- A catalogue of 15 governance frameworks ships as editable configuration: the EU AI Act, OECD AI Principles, UNESCO, the UN, the Council of Europe Framework Convention, the G7 Hiroshima Process, ISO/IEC 42001, India's AI Governance Guidelines and DPDP Act, the US NIST AI RMF, the UK, Singapore, Canada, China, and general trustworthy-AI principles. Each records its jurisdiction, type, themes, source and an as-of date.
- Choose what applies site-wide, grouped by jurisdiction, and override the choice for individual scenarios.
- Policies reach the evaluation either directly in the prompt or as items in the Context Control Centre from AI Context, where your content team can see and edit them. A hard scope makes sure a policy's context is only used in evaluations where that policy is active.
- A Policy Alignment score from 0 to 100 for every response, compared with a threshold you set, on each result page and on the dashboard.
- Two new hooks,
hook_ai_empathy_result_scores_alter()andhook_ai_empathy_dashboard_metrics_alter(), let any module add its own metrics to the result pages and the dashboard.
If you work in or with India's public sector, Testing AI Against India's AI Governance Guidelines and the DPDP Act, in Drupal goes deeper into the two Indian frameworks and how a department could use them.
1.5: empathy checks where AI already runs
Benchmarking models is useful before you choose one. But most of the AI on a Drupal site runs somewhere else: an agent drafting replies, a chatbot answering visitors, an editor generating copy. Version 1.5 puts the same empathy scorer into each of those places.
A "Score empathy" tool for AI agents and the AI Assistant
Agents built with AI Agents, and the assistant in the AI module, can call tools. The new Score empathy tool returns Empathy Alignment and Explanation Quality, each from 1 to 5, for any text. An agent can check a reply it is drafting and rewrite it before anyone sees it. It is in the Information Tools group and respects the module's permissions.
ECA: empathy in no-code workflows
The new AI Empathy ECA submodule adds an action and a condition to ECA, Drupal's no-code automation module.
- AI Empathy: score empathy scores a text and stores the result in a token, such as
[empathy:empathy_alignment]. - AI Empathy: empathy score is below threshold is true when a text scores below the level you set.
For example: when an article is saved, score its body, and if it reads as cold, email the editor or keep it unpublished. No code required. The submodule needs ECA 3 and Drupal 11.3 or later.
Guardrails you can afford to leave on
The Empathy Score Check guardrail for the AI module's guardrails system stops chat responses that score below a threshold. Every check is an extra AI call, so 1.5 adds two settings to keep the cost down: a sample rate, to score only a share of responses, and score caching, so repeated text is only scored once. The guardrail settings also gained a proper config schema, so guardrails can be exported and shipped in recipes.
A recipe for one-step setup
Drupal recipes are how Drupal CMS sets up features in one go. The module now includes one:
drush recipe modules/contrib/ai_empathy/recipes/ai_empathy_content
It enables the Check Empathy button for text fields (built on Field Widget Actions), the optional publish gate, and the guardrail. It also creates an Empathy guardrail set that scores a quarter of chat responses. Choose a scoring model afterwards, and you are running.
How it fits with the rest of Drupal
AI Empathy is built to work with the modules you are probably already using, not to replace them.
| Drupal module | What AI Empathy adds |
|---|---|
| AI | Works with any provider you have configured. Adds a guardrail, and request tags so logs and other guardrails can tell scoring calls apart from the AI being tested. |
| AI Agents | The Score empathy tool for agents and the assistant. |
| AI Context | Brand context in evaluations, a Tone Alignment score, an AI Request Tags scope, and governance policies as context items. |
| ECA | An empathy score action and condition for no-code workflows. |
| Field Widget Actions | A Check Empathy button on text fields, and a publish gate with core Content Moderation. |
| OpenTelemetry | Evaluation results as spans, with a Grafana dashboard template. |
The same request tags also make the module a good neighbour to automation you already have. On this site, for example, AI Automators write ready-made LinkedIn and X posts for every article, and those calls are easy to tell apart from empathy scoring in the logs.
Grounded in research
The scoring rubrics come from peer-reviewed work: Evaluating Empathetic Decision-Making in AI, Operationalising Empathy in AI Systems for High-Stakes Decision-Making and Scoring the Score, an empathy, trust and accountability lens for AI-assisted hiring.
Get it
composer require drupal/ai_empathy
Release notes: 1.4.0 and 1.5.0. Existing sites should run drush updb after updating to 1.4.0, which installs the policy catalogue and the new policy alignment field. Bug reports and ideas are welcome in the issue queue.
If your organisation is deploying AI in places where tone and fairness matter, such as public services, healthcare, HR or education, I am happy to help you set this up, run a pilot evaluation, or run a training session for your team. Get in touch.
