Algorithms draft the shift plan, dashboards score productivity, and an HR tool has quietly flagged you as a „low performer”. German law has answers that most employees — and quite a few employers — do not know: AI may inform personnel decisions in Germany, but it may not make them. Fully automated dismissal decisions are barred by Art. 22 GDPR, monitoring-capable AI needs works-council agreement under § 87 BetrVG, and AI-generated „performance data” can be attacked in court — often effectively, because employers struggle to explain their own tools.
Algorithmic management and co-determination (§ 87 BetrVG)
Algorithmic management is everyday reality: automated shift and route planning, productivity scoring, workforce analytics, attrition prediction, AI-assisted reviews. Every one of these tools runs into the works council’s co-determination rights:
- § 87(1) Nr. 6 BetrVG — technical systems objectively capable of monitoring behaviour or performance may not be introduced or operated without works-council agreement. The BAG applies this broadly: capability is enough, and AI scoring tools are monitoring technology almost by definition.
- § 87(1) Nr. 2 and 3 BetrVG — algorithmic scheduling touches the distribution of working time, another mandatory co-determination matter.
- § 80(3) BetrVG — when assessing AI systems, the works council may bring in an external expert, a right the legislator made explicit precisely for AI.
- § 90 BetrVG — the employer must inform and consult the works council when planning the deployment of AI in work processes.
An AI tool rolled out past the Betriebsrat is operating unlawfully, and the works council can force it offline. If you are being evaluated by software, your first question should be: which Betriebsvereinbarung covers this tool, and what does it allow? See our overview of works council basics for how to raise this.
Art. 22 GDPR: no fully automated dismissal
Art. 22 GDPR gives you the right not to be subjected to a decision based solely on automated processing that produces legal effects for you — and a dismissal is the paradigm legal effect. The consequences:
- A dismissal „decided” by an algorithm and merely executed by HR is unlawful. A human must genuinely make the decision — meaningful involvement, not a rubber stamp on whatever the ranking tool produced.
- The CJEU reads „decision” broadly: even a score can qualify where others draw on it in a determining way. Where the line runs for workplace scoring that feeds into dismissals is still being worked out — an open flank for the defence.
- Where automated processing including profiling is used, you are entitled to meaningful information about the logic involved (Art. 15(1)(h) GDPR) — a disclosure lever that pairs naturally with the transparency attacks below.
AI-generated performance data in dismissal cases
Here is what employers overlook: in a dismissal-protection case, the employer bears the full burden of proving the dismissal grounds. An algorithmic score is not a fact — it is the output of a system whose workings the employer must be able to explain. We attack AI-based dismissal grounds along four lines:
- Transparency: what exactly does the score measure, with which criteria, weights and data basis? Employers frequently cannot answer, because the vendor treats the model as a trade secret. A ground the employer cannot substantiate is a ground the court cannot accept.
- Data quality: misattributed activity, uncounted task types, systems that penalise precisely the complex work the employee handled — error patterns are the rule, not the exception.
- Benchmark validity: for a performance-based dismissal, the case law demands substantial, sustained underperformance measured against a valid comparison group — a percentile rank in a dashboard does not establish that on its own.
- Procedural defects: no Betriebsvereinbarung covering the tool, no proper works-council consultation on the dismissal, no prior warning. Each defect independently weakens or voids the dismissal.
Employee use of ChatGPT & co.
The other direction matters just as much: employees using generative AI at work.
Policy violations as a conduct issue
Whether you may use external AI tools is governed by your employer’s instructions and IT policies. Using a prohibited tool is a conduct issue, so the ordinary proportionality ladder applies: for a first violation without serious harm, an Abmahnung must precede any dismissal under the framework of conduct-based dismissal. Where no clear policy existed at all, even a warning stands on shaky ground — an employee cannot violate a rule that was never communicated.
The serious case: confidential data in the prompt
The genuinely dangerous scenario is feeding confidential material into external AI systems — source code, customer personal data, financials, trade secrets. That can breach confidentiality duties, the GDPR and trade-secret protection under the GeschGehG regardless of any AI policy — in grave cases supporting extraordinary dismissal. The severity depends on sensitivity, scope, intent and harm — but this is the category where „I was just being efficient” stops working as a defence. If it has already happened: do not delete traces in panic, and get advice before the Personalgespräch.
AI policies and the works council
Sensible employers regulate AI use in a Betriebsvereinbarung — which tools are approved, what data may be entered, how AI-assisted evaluations may be used. For employees this is protective in both directions: clear rules on what you may use, and negotiated limits on how you may be measured. Where guidelines for AI-assisted personnel decisions are set up, the works council’s co-determination rights apply — worth asking your Betriebsrat what exists before a conflict arises.
The EU AI Act at the workplace
The EU AI Act adds a product-regulation layer on top of the GDPR and the BetrVG. The workplace-relevant core:
- High-risk classification: AI systems used for recruitment, for promotion and termination decisions, for task allocation based on behaviour or personal traits, and for monitoring and evaluating performance are classified as high-risk. That triggers duties for providers and for employers as deployers: risk management, logging, human oversight, and information duties — including informing employees and their representatives before such systems are used at the workplace.
- Outright bans: emotion-recognition systems at work are prohibited, save for narrow medical and safety purposes — the „mood scanner” in your video calls is not a compliance grey zone; it is banned.
- Timing: the AI Act’s obligations phase in on a staggered statutory timetable, and supervisory guidance is still concretising how the workplace rules will be enforced — verify specific compliance-date questions case by case rather than assuming them.
For an individual dismissal dispute, the AI Act rarely decides the case on its own — but a high-risk tool deployed without the required oversight and information duties is one more documented defect in the employer’s position.
Worked example
Customer-service agent in Cologne, €3,800 gross monthly, 5 years of service. An AI quality tool scores all calls; the employer dismisses her with the ordinary notice period for „sustained underperformance”, citing three months in the bottom decile of the team dashboard. No prior Abmahnung; the tool was introduced without any works-council agreement although a Betriebsrat exists.
We file the Kündigungsschutzklage and pair it with an Art. 15 GDPR request including the Art. 15(1)(h) logic disclosure. The response reveals the scoring heavily penalises call duration — and our client staffed the escalation queue for complex cases, which the tool did not segment. The employer cannot explain the model’s weighting in its pleadings, cannot show a valid comparison group, and concedes the missing Betriebsvereinbarung.
Settlement at the Gütetermin: dismissal converted into a termination by mutual agreement, severance of 1.0 monthly salaries per year of service (€19,000) — against an opening offer of €7,600 —, the 3-month notice period paid as garden leave (€11,400), a Grade-2 reference, and deletion of the score-based performance entries from her file. Total package: approximately €30,400.
Frequently asked questions
Can I be dismissed by an algorithm?
Not lawfully. Art. 22 GDPR bars decisions with legal effect based solely on automated processing, and dismissal is the textbook example. A human must genuinely make and own the decision — and in litigation, explain the grounds without hiding behind the tool.
My employer uses a productivity score. Can I see how it works?
You can demand your data and meaningful information about the logic of automated processing via Art. 15 GDPR, the works council can demand full system transparency under the BetrVG, and for high-risk systems the AI Act adds information duties towards the workforce.
Am I allowed to use ChatGPT for work tasks?
Check your employer’s policy first — and regardless of policy, never enter customer data, colleagues’ personal data, code or confidential business information into an unapproved external tool. Policy questions are an Abmahnung topic; confidential-data uploads can be a dismissal topic.
I was warned for using an AI tool no policy ever prohibited. Is that valid?
Questionable. A warning presupposes a breach of duty, and without a communicated rule or an obvious confidentiality issue there is often none. Unjustified warnings can be challenged and removed from your personnel file.
Does the EU AI Act give me a personal claim against my employer?
Its duties are mainly enforced by supervisory authorities, and it includes a right to explanation for certain high-risk decisions; how far individuals can rely on it directly is still being clarified. In practice we use AI Act violations as supporting defects alongside the GDPR and BetrVG arguments, which carry the case.