Legal certainty

Legal certainty

The blocker in most boardrooms is not the technology. It is who answers for what the AI said.

An AI agent is not a legal person. Under Swiss and EU law it is a tool, and a company is bound by what its tools declare in much the same way a bank is bound by its cash machine. That is a settled principle, and it is exactly why the guardrails matter more than the model.

Most discussion of this jumps to smart contracts and automated price corridors. Useful eventually, but it skips the question a lawyer actually asks first: can you show what was said, to whom, on what basis, and who authorised it?

That is an evidence problem before it is an automation problem — and it is solvable today, which is why this product solves that end first.

What this looks like in practice

Binding is off until you turn it on

By default the twin never promises a price, a date or a term. It gives indicative guidance and defers to your team. Nothing can be logged as binding unless the owner explicitly enabled it.

A person signs, and the signature carries proof

Contracts are signed by a named signatory with an explicit consent step. The server records the timestamp, IP address, browser and the signer's email — none of it accepted from the client.

The text is frozen at signature

The contract body is hashed with SHA-256 and the signature binds to that hash. The countersigning party sends back the hash it was shown, so nobody can sign a document other than the one they read, and terms cannot be edited from under a signature.

Everything said is on the record

The statement register keeps every answer the twin gave the outside world, timestamped and source-attributed, and exports as JSON for your files.

What is true today

This section exists because most pages in this category do not have one. Both columns are maintained against the product's own status list, not written by marketing.

Works today
  • Commitment levels with binding statements off by default
  • Electronic signature with name, consent, timestamp, IP and user agent
  • Terms frozen by SHA-256; countersignature verifies the same hash
  • Statement register, exportable
  • Swiss law and jurisdiction stated in the framework agreement
Does not, and we will not pretend
  • This is a simple electronic signature, not a qualified one (QES). For Swiss and EU service contracts that is generally sufficient, but contracts requiring statutory form are out of scope.
  • There are no automated price corridors or spending limits an agent may transact within.
  • The contract templates are written for the platform's own use and have not been reviewed by external counsel. Have yours read them before you rely on them at scale.
See the full status of every feature

Check it against your own website

The free Twin-Check reads your site the way an AI assistant would and tells you what it can and cannot answer about you. No account needed.

Questions people actually ask

Is a signature made here legally valid?

It is a simple electronic signature: explicit consent plus the signatory's typed name, with timestamp, IP and a hash of the exact text recorded. For service agreements under Swiss and EU law that is generally valid, since no particular form is prescribed. Contracts with a statutory form requirement need a qualified signature, which this is not.

What if our AI says something we did not authorise?

Binding statements are off unless you enabled them, so by default nothing it says is presented as a commitment. The statement register lets you show precisely what was said and on what sources.

Can we set spending limits for the agent?

Not today, and we would rather say so than imply otherwise. The agent prepares; a person signs. Automated limits only make sense once agents transact unattended, which is not where the technology honestly is.

This page answers
ai contract liabilitywho is liable ai agent contractelectronic signature switzerland b2bai negotiation legal riskbinding ai statements company