Section 22 AGG: Reversal of the burden of proof in discrimination cases — what employers need to know

1. Wording of § 22 AGG

Short answer: § 22 AGG allocates the burden of proof in two stages — the claiming party has to prove indications suggesting a disadvantage on one of the grounds named in § 1 AGG; if it succeeds, the other party bears the burden of proving that there was no breach.

"Where in a dispute one party establishes indications suggesting a disadvantage on one of the grounds named in § 1, the other party bears the burden of proving that there has been no breach of the provisions protecting against disadvantage." — § 22 AGG

Allocation of the burden of proof:

• Applicant: indications (low threshold)
• Employer: full proof of non-discrimination (high threshold)

EU-law background: Art. 19 of Directive 2006/54/EC (gender, employment), Art. 10 of Directive 2000/78/EC (employment framework) and Art. 8 of Directive 2000/43/EC (anti-racism) require the defendant to provide counter-evidence as soon as the person concerned establishes facts from which direct or indirect discrimination may be presumed.

The grounds to which § 22 AGG attaches are listed in § 1 AGG: race or ethnic origin, gender, religion or belief, disability, age and sexual identity. The prohibition of disadvantage in § 7(1) AGG also applies where the acting person merely assumes such a ground to be present; agreements breaching it are void under § 7(2) AGG. § 22 AGG applies only in Germany — for Austria see section 7.

2. What are 'indications'?

Short answer: indications are facts suggesting a disadvantage on a characteristic under § 1 AGG — the wording of a job advertisement, conspicuous selection rates, statements made in an interview, temporal links or documented bias findings of a selection tool.

BAG case law recognises the following as indications:

• Wording of job advertisements (e.g. "young, dynamic team") — the advertisement is the point of reference that is easiest to evidence because it is publicly documented; § 11 AGG expressly prohibits advertising a post in breach of § 7(1) AGG
• Statistical anomalies in selection procedures — see section 4
• Discriminatory statements in the job interview (recording, witnesses) — harassment, too, is a separate form of disadvantage under § 3(3) and (4) AGG, not merely an accompanying circumstance
• Temporal links (dismissal after pregnancy) — where the measure is a reaction to someone having exercised rights under the AGG, the prohibition of victimisation in § 16(1) AGG applies in addition
• AI tool outputs with documented discrimination — see section 4

When assessing this, it is necessary to distinguish which form of disadvantage is at issue: direct disadvantage exists where a person receives less favourable treatment on a ground under § 1 AGG (§ 3(1) AGG); indirect disadvantage exists where apparently neutral provisions, criteria or procedures are liable to put persons at a particular disadvantage (§ 3(2) AGG). With selection criteria, filter rules and algorithms it is typically indirect disadvantage that is at issue — because there nobody names the protected characteristic, yet it takes effect through correlating criteria.

3. Rebuttal by the employer

Short answer: once the claiming party has produced indications, the employer must provide full proof of non-discrimination — typically through a documented AGG-compliant job advertisement, selection criteria defined in writing in advance, objective assessment sheets per applicant, attendance lists for the interviews and a written, reasoned selection decision. With AI-supported selection, additionally: bias test results and audit logs.

§ 22 AGG governs the burden of proof, not the means of proof — which documents are persuasive in an individual case is for the court to decide. The following list is the practice recommendation with which this counter-evidence can be provided at all; its legal anchor is § 12(1) AGG, which obliges the employer to take the necessary protective measures, including preventive ones.

Requirements for full proof:

1. Job advertisement documented — neutrally worded, AGG-compliant. The text of the advertisement is regularly the first indication and therefore also the first piece of exculpatory evidence (§ 11 AGG).
2. Selection criteria defined — in writing before applications are received. Only a criteria grid fixed in advance shows that the decision was not justified after the event.
3. Assessment sheets per applicant — objective point scores. They make the comparison between applicants traceable and reveal indirect disadvantages under § 3(2) AGG earlier.
4. Attendance list for interviews — who was present, when and with whom. It determines which witnesses can be heard at all on an alleged statement.
5. Reasons for the selection decision — a written note, dated. It supports the assertion that a different criterion, unrelated to the characteristic, was decisive.
6. With an AI tool: bias test results and logs — for high-risk AI, retaining the automatically generated logs becomes a deployer obligation from 2 December 2027 (Art. 26(6) in conjunction with Art. 12 AI Act, start of application for Annex III systems under Art. 113(3)(c)(i) AI Act as amended by Regulation (EU) 2026/1744, see section 4).

4. Statistical and algorithmic indications: AI in the selection procedure

Short answer: statistical anomalies in selection procedures are recognised as an indication within the meaning of § 22 AGG — in 8 AZR 488/19 (21 January 2021) the BAG accepted the median comparative pay as an indication. That the indication logic also applies to algorithmic selection outputs is the obvious continuation of this line, but not a decided case: a BAG decision specifically on algorithm outputs is not among the judgments evaluated here (see section 5). The indication threshold is low:

• Significant deviations in rates along protected characteristics (age, gender, origin, disability)
• Systematic filtering out of a comparison group (e.g. applicants > 50)
• Black-box models without audit or logic information (cf. CJEU C-203/22 on the scope of the right to information)
• Correlated proxy variables without documented mitigation — that is, substitute criteria close to a characteristic, such as postcode, club membership or an unbroken employment history, which can correlate strongly with protected characteristics; 'digital native' filters and similar feature scores often operate as age filters. This is the typical case of indirect disadvantage under § 3(2) AGG: the protected characteristic is never named, but takes effect through the substitute criterion. Review step (practice recommendation): go through the tool's list of criteria and, for each criterion, state in writing why it is objectively necessary for the post; otherwise remove the criterion.
• Rejection by the algorithm alone, without human review (Art. 14, 26 AI Act — start of application for Annex III systems 2 December 2027 under Art. 113(3)(c)(i) AI Act as amended by Regulation (EU) 2026/1744)

BAG 8 AZR 300/24 (23 October 2025, Daimler Truck) shows the trend: there, even a single comparison case sufficed as an indication in the equal pay context — a documented bias finding of a selection algorithm should, on that basis, carry all the more weight as an indication.

Anyone using an AI system for applicant selection is a deployer within the meaning of the AI Act, Regulation (EU) 2024/1689. Annex III no. 4 classes AI systems for recruitment and selection — in particular for placing targeted job advertisements, for screening and filtering applications and for evaluating candidates — as high-risk systems. Under Art. 113, the Regulation has in principle applied since 2 August 2026. With Regulation (EU) 2026/1744 (Digital Omnibus AI), in force since 27 July 2026, the high-risk obligations of Chapter III Sections 1 to 3 apply to Annex III systems — and thus to recruiting AI under Annex III no. 4 — only from 2 December 2027, and to Annex I systems from 2 August 2028; the general start of application of 2 August 2026 remains unchanged.

The following overview states, for each obligation, the trigger to which the provision attaches it — not every line applies to every employer. The obligations under Art. 26 and Art. 27 AI Act apply to Annex III systems from 2 December 2027 under Art. 113(3)(c)(i) AI Act as amended by Regulation (EU) 2026/1744; the right to information under Art. 15(1)(h) GDPR applies irrespective of that already:

Breaches of the deployer obligations under Art. 26 AI Act are subject, under Art. 99(4) AI Act, to fines of up to EUR 15,000,000 or, in the case of undertakings, up to 3 % of total worldwide annual turnover for the preceding financial year — whichever is higher; for SMEs including start-ups and, since 27 July 2026, for small mid-cap enterprises, the lower of the two amounts applies instead (Art. 99(6) and (6a) AI Act). As an EU regulation, the AI Act applies equally in Germany and Austria; there is no difference between the countries here.

What a bias review means in practice. Its core is the statistical evaluation of the algorithm's previous decisions by protected characteristics, with documented mitigation — before productive use and repeatedly thereafter. The rhythm and the detailed arrangements do not follow from § 22 AGG, which only allocates the burden of proof; the following four steps are therefore a practice recommendation unless a reference is expressly named:

1. Form comparison groups along the grounds of § 1 AGG and set the pass-through rates for each stage of the procedure — pre-selection, invitation, offer — against each other. Only the stage-by-stage comparison shows at which point a group drops out.
2. Record anomalies with the date, the data status and the person carrying out the review. Without these three details it cannot be evidenced later what a result related to.
3. Place the root cause analysis and the measure taken in the same record and retain them together with the system's automatically generated logs — retaining those for at least six months becomes a deployer obligation for high-risk AI from 2 December 2027 (Art. 26(6) in conjunction with Art. 12 AI Act, start of application under Art. 113(3)(c)(i) AI Act as amended by Regulation (EU) 2026/1744).
4. Name the responsibility — obviously the trained person who exercises human oversight of the system under Art. 26(2) in conjunction with Art. 14 AI Act.

5. Key judgments (BAG and CJEU)

Short answer: the reversal of the burden of proof under § 22 AGG is shaped by central BAG judgments: BAG 8 AZR 300/24 (23 October 2025, Daimler Truck) — a single comparator colleague suffices as an indication; BAG 8 AZR 488/19 (21 January 2021) — median pay as a statistical indication; BAG 8 AZR 450/21 (16 February 2023) — 'better negotiating skills' is not sufficient as a justification; BAG 8 AZR 21/24 (19 September 2024) — no compensation for an abusive application; BAG 8 AZR 209/21 (8 May 2025) — transfer of employee data within a group to an HR system; CJEU C-203/22 (27 February 2025) — right to information about the logic of automated decisions.

The table deliberately does not name amounts: the level of compensation depends on the individual case; the statutory limits and minimum amounts for Germany and Austria are set out in section 8.

Three of the BAG decisions named are pay cases — which is why the right to information on comparative pay is a practically important source of indications (see section 9). BAG 8 AZR 300/24 is an equal pay judgment and not an algorithm judgment; the transfer to algorithmic selection is to be read as a trend, not as one-to-one evidence.

Two rows of the table concern not the indications themselves but their margins. BAG 8 AZR 21/24 (19 September 2024) limits the other side: applications without any intention to take up employment are not protected, as an abuse of rights, and compensation is ruled out — for employers this means securing the application history and the communication (practice recommendation). BAG 8 AZR 209/21 (8 May 2025) concerns the transfer of employee data within a group to an HR system and shows that, alongside claims under the AGG, a separate claim for damages under Art. 82 GDPR may exist depending on the facts; clarifying the legal basis and the circle of recipients before the rollout is the resulting practice recommendation.

In C-203/22 (Dun & Bradstreet Austria) the CJEU set out the scope of Art. 15(1)(h) GDPR: in the case of automated individual decisions within the meaning of Art. 22 GDPR, data subjects are entitled to meaningful information about the logic involved — in such a way that the decision can be understood and reviewed. A blanket reliance on trade secrets is not sufficient; in the event of a conflict, the information is to be disclosed to the competent authority or the court, which weighs the interest in confidentiality against the interest in information. Practical consequence: anyone buying selection software needs audit and information rights in the contract, otherwise they can neither explain the logic nor evidence it in proceedings (practice recommendation).

6. Practice guide

Short answer: eight practical steps minimise the § 22 AGG risk: a job advertisement audit with AGG-compliant language, selection criteria fixed in writing before applications are received, assessment sheets per applicant, with AI tools a bias review as a practice recommendation (CNIL guidelines; the processing of special categories of data for bias detection is exceptionally permitted by Art. 4a AI Act — inserted by Regulation (EU) 2026/1744 and applicable since 27 July 2026), annual HR training, anonymised first-round selection, structured interviews with a score sheet and retention of the documentation for 3 years.

1. Job advertisement audit with AGG-compliant language — the standard of review is § 11 AGG (advertising without breaching § 7(1)); in Austria, § 9 GlBG requires gender-neutral advertising, and a breach is an administrative penalty of up to EUR 360 imposed by the district administrative authority under § 10 GlBG
2. Selection criteria in writing BEFORE the application (practice recommendation) — they are the backbone of the counter-evidence under § 22 AGG
3. Assessment sheets per applicant (practice recommendation) — objective point scores instead of free text
4. With AI: bias review as a practice recommendation (CNIL guidelines; the permission in Art. 4a AI Act for processing special categories of data for bias detection has applied since 27 July 2026, following Regulation (EU) 2026/1744)
5. Annual training of the HR team — the rhythm is a practice recommendation; the legal effect of training under § 12(2) AGG is explained in section 10
6. Anonymised first-round selection (removing photo, name, date of birth) (practice recommendation) — it removes the basis for the typical carriers of indications
7. Structured interviews with a score sheet (practice recommendation) — the same questions for all applicants make the comparison robust
8. Retain the documentation — three years is a practice recommendation (limitation periods for claims plus a reserve); the relevant deadlines for Germany and Austria are set out in section 8

7. Austria: § 12(12) and § 26(12) GlBG instead of § 22 AGG

Short answer: § 22 AGG does not apply in Austria; its place is taken by the rule on the burden of proof in the Austrian Equal Treatment Act — § 12(12) GlBG for discrimination on grounds of gender (Part I) and § 26(12) GlBG for the other grounds (Part II). The standard is a different one, not merely a different section number.

Two differences are decisive:

• Entry threshold: under § 22 AGG, the claiming party has to prove the indications. Under § 12(12) and § 26(12) GlBG it is sufficient for the person concerned to show the elements of discrimination to be credible.
• Standard for exculpation: in Germany the other party must prove that there was no breach of the provisions protecting against disadvantage. In Austria, exculpation already succeeds with proof that, weighing up all the circumstances, it is more likely that another motive was decisive for the different treatment or that a ground of justification applies. Austria therefore works with a weighing of probabilities, Germany with counter-evidence.

The range of protected grounds does not coincide either: the GlBG covers gender (§ 3) as well as ethnic affiliation, religion or belief, age and sexual orientation (§ 17). Disability is not a GlBG ground — for discrimination on grounds of disability in working life, Austria applies the Disability Employment Act (BEinstG, §§ 7a et seq., in particular § 7b), and outside working life the Federal Disability Equality Act (BGStG). Claims are to be asserted before the labour and social court; advice and support are provided by the Ombud for Equal Treatment (GAW, § 3 GBK/GAW Act) and the Equal Treatment Commission (Gleichbehandlungskommission) — not by the German Federal Anti-Discrimination Agency (Antidiskriminierungsstelle des Bundes).

8. Deadlines and compensation compared, Germany and Austria

Short answer: the legal consequences diverge more strongly than the rule on the burden of proof itself — Germany caps compensation where a person is not hired, Austria works with minimum amounts; Germany requires claims under § 15(1) or (2) AGG to be asserted in writing within two months, Austria requires claims for the failure to establish an employment relationship or for a promotion not granted to be asserted in court within six months of the rejection; different deadlines apply there for harassment and for challenging a dismissal (see the table).

The following table reproduces statutory information only — deadlines, amount limits and references to provisions; it contains no practice recommendations:

In addition, the following applies in Austria: a dismissal or termination must be challenged within 14 days (§ 15 and § 29 GlBG); for pay and property claims the three-year limitation period in § 1486 ABGB applies. For harassment the GlBG provides for a minimum amount of EUR 400, and for sexual harassment under § 12(11) GlBG at least EUR 720. The EU-law background to the German cap is Art. 18 of Directive 2006/54/EC: an upper limit is permissible only where the damage consists solely in the refusal to take the application into consideration. The range of one to three gross monthly salaries mentioned in the TL;DR describes BAG practice, not a statutory amount.

9. Pay indications: rights to information as a trigger

Short answer: pay data are a central route by which employees obtain indications within the meaning of § 22 AGG — in Germany through the right to information under §§ 10 to 12 EntgTranspG, whose result (the median comparative pay) delivers precisely the figure the BAG accepted as an indication in 8 AZR 488/19.

Legal framework in Germany: § 7 EntgTranspG prohibits agreeing or paying lower pay on grounds of gender for the same work or work of equal value; § 3 EntgTranspG prohibits direct and indirect pay disadvantage with regard to all components of pay. Employees may request, in text form, information on the average monthly comparative pay (statistical median) for a named comparable activity and on up to two components of pay. The right exists in establishments with as a rule more than 200 employees; the comparative pay is not to be stated where the comparable activity is performed by fewer than six employees of the other gender.

In Austria, § 11 GlBG requires that classification rules at establishment level and norms of collective legal regulation observe, in their pay criteria, the principle of equal pay for the same work or work of equal value; § 3 no. 2 GlBG prohibits discrimination in setting pay.

At EU level, Art. 18 of Directive (EU) 2023/970 (pay transparency) provides for its own reversal of the burden of proof; the transposition deadline ran until 7 June 2026 under Art. 34(1). Important for the classification: under Art. 288(3) TFEU a directive is binding only on the Member States and has no direct effect between private employers and employees (CJEU, judgment of 14 July 1994, C-91/92 Faccini Dori). For private employers, the EntgTranspG and the AGG continue to apply until the national transposition act; an interpretation of the applicable law in conformity with the directive remains possible. The deadline has passed: in Germany, as at 17 August 2026, not even a ministerial draft was available (BT-Drs. 21/7026 of 8 July 2026). In Austria there is neither a ministerial draft nor a government bill at federal level on the pay transparency directive; the 2026 GlBG consultation draft concerns equal treatment in goods and services, not pay transparency. You should check the current state of transposition before applying this.

10. Before the dispute: duties of protection, complaints body, prohibition of victimisation

Short answer: the evidence position arises before the proceedings — § 12 AGG obliges the employer to take protective measures including preventive ones, § 13 AGG gives employees a right to complain, § 16 AGG prohibits victimising those who exercise their rights.

In detail: under § 12(1) to (3) AGG the employer must take the necessary measures to protect against disadvantage, point out that such conduct is unlawful, and in the event of breaches take appropriate measures such as a warning, redeployment, transfer or dismissal; appropriate training counts under § 12(2) AGG as fulfilment of that duty. The Act and the information about the body responsible for complaints must be published in the establishment under § 12(5) AGG. Employees who feel disadvantaged on a ground under § 1 AGG may complain to the competent bodies of the establishment under § 13(1) AGG; the complaint must be examined and the outcome communicated. A properly conducted complaints procedure is thus at the same time documentation for a § 22 AGG case (practice recommendation).

§ 16(1) AGG prohibits disadvantaging employees because they have exercised their rights or refused to carry out an instruction breaching the AGG; the protection also covers persons who provide support or give evidence as witnesses. Austria has the same structure in § 13 GlBG (Part I) and § 27 GlBG (Part II): nobody may be dismissed, given notice or otherwise disadvantaged as a reaction to an internal complaint or to the initiation of proceedings — witnesses and persons providing information included. The EU-law basis is Art. 24 of Directive 2006/54/EC and Art. 11 of Directive 2000/78/EC; the EU-law reference point for prevention is Art. 26 of Directive 2006/54/EC, which asks the Member States to encourage employers to take effective preventive measures.