sebbi.pro
Checks / AI Defence File
EU Product Liability Directive · in force by 9 December 2026

If you can't prove what your AI did, the court can assume it was faulty.

From 9 December 2026 the new EU Product Liability Directive (2024/2853) treats software and AI as products. People harmed by a defective product can claim without proving fault, a court can order you to disclose evidence, and if you can't, or the case is technically complex, it can presume your AI was defective. Check how ready your evidence is, and leave with a sealed AI Defence File.

PLD
software and AI products in scope identified
A note of which of your products are software or contain AI, so are now 'products' under the Directive.
PLD
inventory of AI and software products on the EU market
A list of every AI or software product you place on the EU market, with versions.
PLD
version history kept for every release
Release notes for every version, so you can show which version caused or did not cause harm.
PLD
security update policy for the product's expected life
A written policy for security updates over the product's expected life.
PLD
automatic logs of what the AI did in use
Automatic logging of what the AI did, switched on in the product itself.
PLD
logs are tamper-evident and stand up as evidence
Records sealed so nobody, including you, can edit them later. Courts trust evidence that can be checked.
PLD Art.9
procedure to disclose evidence fast on a court order
Who gathers and hands over evidence if a court orders disclosure, and how fast.
PLD
pre-release safety testing results kept
Test results from before each release, kept with the version they tested.
PLD
controls on changes after release (updates, self-learning)
How changes after release (updates, self-learning) are approved and recorded.
PLD
incident and complaint log
A log of incidents and complaints and what you did about each.
PLD
modifications reviewed (a substantial modification makes you the manufacturer)
A check on modifications: a substantial change can make you the manufacturer.
PLD
supplier contracts cover liability and evidence sharing
Contracts with model and component suppliers that cover liability and sharing evidence.
PLD
cybersecurity measures in place
Security controls and vulnerability handling. A security flaw can make a product defective.
PLD
product liability insurance covers software and AI
Product liability insurance that explicitly covers software and AI.
PLD
records kept >= 10 years (the liability period)
Years you keep the records. 10 or more passes; up to 25 for latent injury claims.

Questions people ask

Does the new Product Liability Directive cover AI and software?

Yes. Directive (EU) 2024/2853 expressly treats software, including AI systems, as a product. Manufacturers can be liable for damage caused by a defective product without the claimant having to prove fault.

When does it apply?

Member States must bring it into national law by 9 December 2026. It applies to products placed on the market or put into service after that date.

What is the disclosure and presumption rule?

A court can order the defendant to disclose relevant evidence. If the defendant fails to, or if proving the defect is excessively difficult because of technical or scientific complexity, the court can presume the product was defective or that the defect caused the damage.

Can a software update make my product defective?

Yes. A product can be defective because of updates, a lack of security updates the manufacturer is responsible for, or the effect of self-learning after it was placed on the market, while it remains within the manufacturer's control.

How long can claims be brought?

In general up to 10 years after the product was placed on the market, extended to 25 years where a personal injury is latent. Keep the evidence at least that long.

How do I defend a claim about my AI?

With records you can produce on demand that show what the AI did, which version was running, what testing it passed and that the records have not been changed. sebbi.pro seals each AI decision so it can be disclosed and checked by anyone.

Every control

Each of these 15 controls has its own page with a live check: AI Defence File checks. From any AI assistant connected to https://sebbi.pro/mcp, run sebbi_run_framework with framework=pld.
Seal every AI decision so you can always disclose it. Start free
AI news
MIT Technology ReviewAI’s hidden storage bill: What inference leaves behind1hTechCrunchBlackstone’s Jas Khaira says fears over AI-buildout debt are overblown10hThe VergeMeta to create a new class of shares to fund AI push12hTechCrunchRecursive launches Flint, a laptop coding model it says was largely built by AI13hThe VergeAnthropic takes on Muse with Claude Duet, its new AI agent13hArs TechnicaThe new human-crafted Apple Music brand matters in an AI world14hTechCrunchWhile Amazon blocks AI agents, Walmart rolls out the welcome mat14hThe VergeWalmart is inviting AI agents to shop for you14hArs TechnicaOpenAI's latest AI models are coming to Google Cloud and Antigravity14hTechCrunchPHP Armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internet15hBBC NewsJust Eat apologises for AI food mistakes after social media ridicule16hArs TechnicaThe $15 billion ad-powered AI startup your doctor may already be using18hThe VergeConversational Gemini navigation comes to Apple CarPlay18hTechCrunchAI promises productivity, but for managers, it’s mostly a headache for now19hWiredAI Guardrails Aren’t Stopping Criminal Hackers19hThe VergeOpenAI says an AI agent lied about disabling its safety monitoring system19hWiredOpenAI Invests in Startup That Uses Brain Implants to Detect Cancer23hArs TechnicaMeta drops the word "train" from its commercial TOS after AI backlash1dArs TechnicaGemini's Personal Intelligence is coming to cars1dBBC NewsOpenAI CEO Sam Altman wins dismissal of sister's lawsuit1dMIT Technology ReviewThe new economics of enterprise AI1dMIT Technology ReviewAI adoption by industry shows what other sectors can learn1dBBC NewsEx-OpenAI worker accused of planting bug had Saudi support, claims prosecutors2dMIT Technology ReviewMaking the internet better for AI agents might make it worse for everyone else2dMIT Technology ReviewAI’s hidden storage bill: What inference leaves behind1hTechCrunchBlackstone’s Jas Khaira says fears over AI-buildout debt are overblown10hThe VergeMeta to create a new class of shares to fund AI push12hTechCrunchRecursive launches Flint, a laptop coding model it says was largely built by AI13hThe VergeAnthropic takes on Muse with Claude Duet, its new AI agent13hArs TechnicaThe new human-crafted Apple Music brand matters in an AI world14hTechCrunchWhile Amazon blocks AI agents, Walmart rolls out the welcome mat14hThe VergeWalmart is inviting AI agents to shop for you14hArs TechnicaOpenAI's latest AI models are coming to Google Cloud and Antigravity14hTechCrunchPHP Armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internet15hBBC NewsJust Eat apologises for AI food mistakes after social media ridicule16hArs TechnicaThe $15 billion ad-powered AI startup your doctor may already be using18hThe VergeConversational Gemini navigation comes to Apple CarPlay18hTechCrunchAI promises productivity, but for managers, it’s mostly a headache for now19hWiredAI Guardrails Aren’t Stopping Criminal Hackers19hThe VergeOpenAI says an AI agent lied about disabling its safety monitoring system19hWiredOpenAI Invests in Startup That Uses Brain Implants to Detect Cancer23hArs TechnicaMeta drops the word "train" from its commercial TOS after AI backlash1dArs TechnicaGemini's Personal Intelligence is coming to cars1dBBC NewsOpenAI CEO Sam Altman wins dismissal of sister's lawsuit1dMIT Technology ReviewThe new economics of enterprise AI1dMIT Technology ReviewAI adoption by industry shows what other sectors can learn1dBBC NewsEx-OpenAI worker accused of planting bug had Saudi support, claims prosecutors2dMIT Technology ReviewMaking the internet better for AI agents might make it worse for everyone else2d