Evidence locked early
Logs, prompts, versions and outputs preserved before updates erase them — the step that decides whether the case can be proved at all.
AI-related disputes
The system produced the harm. Someone still answers for it.
Aun & Co. litigates disputes involving AI systems: vendor and procurement failures, output-based harms, data misuse and responsibility allocation.
AI disputes are commercial disputes with a new evidentiary core: what the system was promised to do, what it actually did, and who controlled the difference. Claims arrive as contract cases against vendors, as tort and defamation cases over harmful outputs, and as data-protection matters over what was fed in. Israeli courts resolve them with existing doctrine — contract, negligence, privacy, consumer law — applied to logs, model behaviour and procurement files. The firm litigates on both sides, informed by running AI inside its own practice.
The work spans
- Vendor disputes over AI systems that failed their promised performance
- Claims and defences over harmful, false or biased AI outputs
- Data-misuse disputes: training data, client data, scraping claims
- Responsibility allocation among vendor, integrator and deploying business
- Evidence strategy for logs, prompts, versions and model behaviour
- The AI system you procured underperforms its contractual benchmarks and the vendor blames your data.
- An automated output about a person or business was false and the demand letter has arrived.
- A vendor trained on data your contract never permitted it to touch.
- Your customer deployed your model outside its documented scope and routes the resulting claims to you.
- A regulator has opened an inquiry into how your organisation used an AI system.
The firm treats the technical record as the pleading's foundation: contracts and SLAs on one axis, logs and version history on the other. The first task in every AI dispute is locking the evidence — systems update, logs rotate, and yesterday's model behaviour is unreproducible next quarter. Liability is then framed through the deployment chain: who selected, who configured, who supervised, who ignored the warning signs.
04 · What you get
The chain mapped
Vendor, integrator, deployer, supervisor: responsibility argued along the actual control points, not the marketing labels.
Technically literate counsel
The firm uses these systems daily; expert evidence is briefed and cross-examined by counsel who understand what the logs say.
A typical engagement: a business deployed a vendor's AI tool for customer-facing decisions, harm followed, and vendor and deployer each point at the other's configuration choices. The firm reconstructs the deployment chain from contracts and logs to establish where control — and therefore responsibility — actually sat.
Described in abbreviated, anonymised form to preserve client confidentiality.
Who is liable when an AI system causes damage?
It depends on control. Courts work along the chain — vendor promises, integrator configuration, deployer supervision — using ordinary contract and negligence doctrine. The party that controlled the failure point, or promised against it, usually answers first.
Can AI-generated content be defamatory under Israeli law?
A false and damaging publication does not stop being one because software produced it. The live questions are attribution and fault of the humans and businesses who deployed and published — analysed under the Defamation Law 1965 framework.
What evidence should be preserved when an AI dispute begins?
Immediately: the contract and SLA, system logs, prompts and outputs at issue, version identifiers, configuration records and internal correspondence about the failure. Model updates destroy reproducibility — preservation on day one is the case.
Can an automated decision be challenged as discriminatory?
Yes; an AI-assisted decision in hiring, credit, or eligibility is judged by its effect, and a discriminatory outcome is actionable whether a human or a model produced it.