Documentation menu
On this page
Article 5v1.2.1pip install litmus-screener

LitmusAI

Screens any AI system description against the EU AI Act Article 5 prohibitions and returns per-category Red / Amber / Clear verdicts with regulatory citations, confidence levels, and remediation guidance. It screens all ten Article 5(1) prohibitions, including points (ba) and (bb). Conservative-by-default: prefers Amber over Clear on ambiguity. UNREVIEWED reference ruleset; signed BYO rulesets supported. Apache 2.0, zero network calls, runs entirely offline. The PyPI distribution is `litmus-screener` (the brand is "LitmusAI"; an unrelated `litmus-ai` package already exists on PyPI).

Quick Start

bashpip install litmus-screener

Needs Python 3.11 or newer.

bash# Quick screen from a free-text description
litmus screen --describe "a chatbot for mental health support for teenagers"

# Or from a structured YAML file
litmus init                    # creates a system.yaml template
litmus screen system.yaml --output report.json

# Bring your own lawyer-signed ruleset
litmus use-ruleset your-firm-ruleset.json
litmus screen system.yaml      # report header now reads "(SIGNED by: ...)"

# CI integration with conventional exit codes (0 / 1 / 2 / 3)
litmus screen system.yaml --fail-on red --output report.json
litmus export report.json -o report.sarif --format sarif

Features

  • 38-rule reference ruleset for Article 5(1) points (a), (b), (ba), (bb), (c), (d), (e), (f), (g) and (h); rules for (ba) and (bb) say that they apply from 2 Dec 2026
  • Constrained expression language: no `eval`, no Python execution at evaluation time
  • Per-rule confidence band (high / medium / low) on every verdict
  • SHA-256 input hash on every report; canonical JSON ordering (RFC 8785)
  • SARIF 2.1.0 output for GitHub Advanced Security / GitLab SAST / Azure DevOps
  • JSON and Markdown exporters
  • Bring-Your-Own-Ruleset with signed `RulesetSignature` block
  • `litmus diff-ruleset` for structural diffs between ruleset versions
  • Zero network calls during screening (CI-enforced via pytest-socket)
  • GitHub Action wrapper at `aiexponent/litmusai/.github/actions/litmusai-screen@v1.2.1`

Regulatory Foundation

Article 5Prohibited AI practicesEnforced · 2 Feb 2025

Read the full pillar: EU AI Act Article 5 explainer →

What the regulation requires

1. The following AI practices shall be prohibited: (a) the placing on the market, the putting into service or the use of an AI system that deploys subliminal techniques beyond a person’s consciousness or purposefully manipulative or deceptive techniques, with the objective, or the effect of materially distorting the behaviour of a person or a group of persons by appreciably impairing their ability to make an informed decision, thereby causing them to take a decision that they would not have otherwise taken in a manner that causes or is reasonably likely to cause that person, another person or group of persons significant harm; (b) the placing on the market, the putting into service or the use of an AI system that exploits any of the vulnerabilities of a natural person or a specific group of persons due to their age, disability or a specific social or economic situation, with the objective, or the effect, of materially distorting the behaviour of that person or a person belonging to that group in a manner that causes or is reasonably likely to cause that person or another person significant harm; (e) the placing on the market, the putting into service for this specific purpose, or the use of AI systems that create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage; (f) the placing on the market, the putting into service for this specific purpose, or the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions, except where the use of the AI system is intended to be put in place or into the market for medical or safety reasons;
5(1)(a)5(1)(b)5(1)(e)5(1)(f)

What you face if you don't comply

Article 5 has been applicable since 2 February 2025 (Art. 113(a)). It has been sanctionable under Chapter XII (Articles 99–100) since 2 August 2025 (Art. 113(b)). Placing or using a prohibited-practice AI system on the EU market today exposes the provider, importer, distributor, or deployer to the highest tier of fines in the regulation: up to €35M or 7% of global annual turnover, whichever is higher, under Article 99(3). Article 5(1) has no opt-out route: once a system meets a prohibition's criteria, including its stated exceptions, no mitigation after the fact makes it lawful. Regulation (EU) 2026/1744 adds two more prohibitions, points (ba) and (bb), which apply from 2 Dec 2026. For providers, they apply only where generating that material is the intended purpose, or a reasonably foreseeable outcome without adequate technical safeguards (Art. 5(1a)). Pre-deployment screening before code is shipped is the only defensible posture.

Up to €35M or 7% of global annual turnover, whichever is higher
Article 99(3) · Penalties

For SMEs, including start-ups, the fine is capped at whichever of the two is lower (Art. 99(6)).

How LitmusAI addresses this

  • 5(1)(a)Detects subliminal-manipulation indicators in system descriptions and outputs; flags when a system materially distorts behaviour against the user's interest
  • 5(1)(b)Pattern-matches vulnerable-population markers (minors, persons with disabilities, persons in vulnerable economic situations) and flags exploitation patterns
  • 5(1)(e)Detects untargeted-facial-image-scraping indicators (web crawl + facial recognition + database creation), the exact pattern Article 5(1)(e) prohibits
  • 5(1)(f)Emits a Red verdict on any system combining emotion inference with workplace or education deployment context (without the medical/safety carve-out)

Source: eur-lex.europa.eu/…/CELEX:32024R1689 · Retrieved

Frequently asked questions

Direct answers to common questions about LitmusAI and Article 5. Regulatory citations reference EUR-Lex CELEX:32024R1689.

What does EU AI Act Article 5 prohibit?
Article 5(1) lists ten prohibited AI practices: the original eight, points (a) to (h), and two added by Regulation (EU) 2026/1744, points (ba) and (bb), which apply from 2 Dec 2026. The original eight: subliminal techniques materially distorting behaviour (5(1)(a)), exploitation of vulnerabilities (5(1)(b)), social scoring of natural persons based on social behaviour or personal characteristics (5(1)(c); note: the final regulation as adopted dropped the "by public authorities" limitation that appeared in the 2021 Commission proposal, so the prohibition applies to any actor), individual criminal-risk assessment based solely on profiling (5(1)(d)), untargeted scraping of facial images for facial-recognition databases (5(1)(e)), emotion inference in workplaces and education (5(1)(f)), biometric categorisation inferring sensitive or protected attributes (5(1)(g)), and real-time remote biometric identification in publicly accessible spaces for law enforcement subject to narrow exceptions (5(1)(h)). The two added: AI systems that generate or manipulate realistic images, video or audio of an identifiable person’s intimate parts or of that person in sexually explicit activity, without that person’s explicit consent (5(1)(ba)), and AI systems that generate or manipulate child sexual abuse material or performances within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, unless a “without right” defence applies under national law (5(1)(bb)). Source: Regulation (EU) 2024/1689 Article 5, as amended by Regulation (EU) 2026/1744.
When did Article 5 become enforceable?
Article 5 has been applicable since 2 February 2025 per Art. 113(a) of the EU AI Act. Sanctionable under Chapter XII (Articles 99–100) since 2 August 2025 per Art. 113(b). Between 2 February and 2 August 2025 the prohibitions applied, but national fines under Art. 99 were not yet attachable. Points (ba) and (bb), added by Regulation (EU) 2026/1744, and Article 5(1a) and (1b) apply from 2 Dec 2026 (Art. 113(a), as amended). Source: Regulation (EU) 2024/1689 Article 113, as amended.
Is LitmusAI a substitute for legal review?
No. LitmusAI produces a screening verdict (Red, Amber, or Clear with confidence levels), not a legal opinion. Final determination of whether a system falls within an Article 5 prohibition requires qualified legal counsel.
What is the UNREVIEWED reference ruleset disclaimer?
The reference ruleset shipped with LitmusAI (ruleset-2024-1689-v1.2 since 1.2.0) was authored and reviewed by an internal AiExponent panel and has not been reviewed by an external EU AI Act lawyer. Every report header and `litmus --version` say so. The status is not tied to a version number: a ruleset counts as REVIEWED only when it carries a signature block, and LitmusAI checks the block for shape, not the signature or the signer. Use the BYO-ruleset path if you need lawyer-signed output today.
Can I use a lawyer-signed Bring-Your-Own ruleset?
Yes. `litmus use-ruleset your-firm-ruleset.json` switches the active ruleset; subsequent reports show "(SIGNED by: …)" in the header. Verification of the signature block is structural: LitmusAI checks its shape, not the signature or the signer. Cryptographic verification is a planned feature.
Does LitmusAI cover every Article 5 prohibition?
Yes, with one limit. It screens all ten Article 5(1) prohibitions, including points (ba) and (bb). Points (ba) and (bb) are declared in the generative_media block of system.yaml; a report run before 2 Dec 2026 says that those two prohibitions apply from that date. Points (a) and (b) are screened from your answers to the influence questions in system.yaml (manipulative, deceptive or subliminal techniques, exploitation of a vulnerability, material distortion of behaviour, significant harm), added in version 1.2.0. In an independent test on 175 worked examples from the European Commission's guidelines (3 Nov 2026), with those questions answered, 6 of the 7 expected prohibitions on those points screened Red and 1 Amber. A file that leaves the questions unanswered falls back to the earlier screen, which missed 6 of 7. The author of the 1.2.0 rules had seen the earlier results, so this result is weaker evidence than the first run. The reference ruleset prefers Amber over Clear on ambiguity, so it gives more false positives.
What is the penalty for Article 5 violations?
Up to €35M or 7% of global annual turnover, whichever is higher, under Article 99(3). That is the highest tier of fines in the EU AI Act. Article 5(1) has no opt-out route: once a system meets a prohibition's criteria, including its stated exceptions, no mitigation after the fact makes it lawful. Regulation (EU) 2026/1744 adds two more prohibitions, points (ba) and (bb), which apply from 2 Dec 2026. For providers, they apply only where generating that material is the intended purpose, or a reasonably foreseeable outcome without adequate technical safeguards (Art. 5(1a)). For SMEs, including start-ups, the fine is capped at whichever of the two is lower (Art. 99(6)).
Does LitmusAI make any network calls during screening?
No. Screening runs offline and needs no API key; LitmusAI 1.2.1 has no networked or LLM mode. Outbound network calls are blocked at CI level via pytest-socket.
Is LitmusAI free?
Yes. Apache 2.0 licensed. No telemetry, no remote calls in default mode, no enterprise tier. The PyPI distribution is `litmus-screener`.
Why is the PyPI package called litmus-screener instead of litmusai?
The PyPI name `litmusai` was unavailable due to PyPI name-similarity rules: an unrelated package called `litmus-ai` already exists on PyPI. The brand is "LitmusAI"; the distribution name is `litmus-screener`. Both names resolve to this same tool through the schema.org `alternateName` declared on the docs page.

Known Limitations

  • Reference ruleset is UNREVIEWED: internally panel-authored, no external EU AI Act lawyer review yet.
  • Conservative-by-default: prefers Amber over Clear on ambiguity. The trade-off is more false positives.
  • A screening is a screening, not a legal certification. Final determination requires qualified counsel.
  • BYO-ruleset signature verification is structural: the signature block is checked for shape, not verified cryptographically, and the signer is not checked. Cryptographic verification is a planned feature.
  • Rules for points (ba) and (bb) read the new generative_media block of system.yaml. A file written for 1.0.x says nothing about generative media, so those two rows read Clear until the block is added.
  • Points (a) and (b) are screened from your answers to the influence questions in system.yaml (manipulative, deceptive or subliminal techniques, exploitation of a vulnerability, material distortion of behaviour, significant harm), added in version 1.2.0. In an independent test on 175 worked examples from the European Commission's guidelines (3 Nov 2026), with those questions answered, 6 of the 7 expected prohibitions on those points screened Red and 1 Amber. A file that leaves the questions unanswered falls back to the earlier screen, which missed 6 of 7. The author of the 1.2.0 rules had seen the earlier results, so this result is weaker evidence than the first run.
  • The scope questions cover the exceptions written in Article 5(1) for points (d) to (h). A claimed exception that needs evidence, such as medical or safety reasons, gives Amber for a reviewer. In the same test, 29 of 113 permitted examples screened Amber and 9 screened Red (32 on version 1.1.3). The Red ones are cases the rules do not model: accidental generation, emergency medical use and red-teaming for (ba) and (bb), proportionality for (c), and emotion recognition used only for a person's own training.
  • Article 5 only. Articles 6 (high-risk classification), 9 (risk management), 13 (transparency) are out of scope. See RiskForge / TransparencyDeck.

For the most current status, see GitHub issues.

Contributing

Contributions are welcome; everything is Apache 2.0 licensed. See the contributing guide and open issues.

License

Licensed under the Apache License 2.0. Not legal advice. Not a notified body.

AI Exponent LLC, a USA-registered company, also operates a separate advisory brand. The tool above is product, not advisory. How the two are kept apart.

Evidence flow

One tool covers one article. The full set covers your audit.

Each AiExponent tool emits a named artefact the next tool reads as input. Browse the full toolchain, from Article 5 screening through Article 72 post-market monitoring.

See all tools →

Content verified 2026-10-25