Deadline · 9 December 2026 · In force

EU Product Liability Directive 2024/2853 — what 9 December 2026 changes if you put your brand on the AI you sell.

The revised EU Product Liability Directive (2024/2853) applies from 9 December 2026. It updates Europe's strict (no-fault) liability regime for defective products that cause personal injury or damage — and it now reaches software and AI systems. If you put your brand on the AI you sell to dental or healthcare clients, this is a date you need to be ready for. Audact's per-call evidence helps you show you acted responsibly.

What changes from 9 December 2026

• AI software is classified as a 'product' under EU product liability law for the first time.
• Strict liability — no need to prove fault — applies to defective AI software that causes damage.
• The 'development risk' defence is restricted: vendors cannot escape liability simply because a defect was undetectable at time of release.
• Deployers (agencies that put AI products into service) can be held liable alongside or instead of the original manufacturer.
• Burden of proof is eased for claimants: courts can presume defectiveness if defendants fail to disclose technical documentation.
• Damages extend to data destruction, psychological harm and property damage — beyond physical injury only.

Who is affected

Any agency that deploys voice AI to EU end-clients is a 'deployer' under the PLD. You may be jointly and severally liable if the AI system causes damage — even if the defect originated in an upstream vendor's model or platform.

If you serve legal-tech, financial-services or other non-personal-injury clients, the personal-injury logic above does not apply in the same way — speak to your counsel about your own exposure.

Why it matters for your agency

Today, when an AI call goes wrong, the natural question is "what did the assistant actually say?" In dental and healthcare contexts — appointment triage, medication reminders, pre-treatment instructions — the difference between what the AI was supposed to say and what it actually said can matter a great deal. From 9 December 2026, as the branded provider, you may be the party that has to answer that question.

How a per-call evidence record helps

Audact produces a per-call, tamper-evident record of exactly what the AI did and did not say. That is the kind of contemporaneous documentation an agency can use to show it acted responsibly and exercised due diligence — for example, that the agent followed the agreed script, made the required disclosures, and stayed within scope. It does not remove, limit, or cap your liability, and it is not a guarantee of any outcome. It gives you a record to stand behind.

• A signed, tamper-evident record of each call — what was said, when, by which agent.
• Evidence that required disclosures were made and the approved script was followed.
• Scope-of-practice and brand-governance checks recorded per interaction.
• Audit-ready exports you can hand to your client, your insurer, or your counsel.

Audact’s evidence-chain technology is patent-pending (UK IPO).