Pre-launch notice. This policy is a working draft and will be aligned with the actual capabilities, launch markets, and provider requirements before use.
The principle
A Coworker may help someone prepare, understand, organize, and carry out legitimate work. It must not be used to obscure responsibility, misrepresent AI as a human, or make harmful work easier to scale.
Prohibited work
You may not use Coworker Systems to:
- break the law, facilitate violence, exploitation, abuse, or credible threats;
- deceive, impersonate, defraud, phish, manipulate, or conceal material AI involvement;
- infringe privacy, confidentiality, intellectual-property, or other legal rights;
- create or distribute malware, bypass access controls, or disrupt systems;
- conduct unlawful surveillance or process highly sensitive information without authority;
- generate discriminatory outcomes or evade rules that protect people from unfair treatment;
- send spam or carry out unwanted mass outreach.
Human review
Do not rely on a Coworker as the sole decision-maker where an error could materially affect a person’s rights, safety, healthcare, employment, housing, education, credit, insurance, legal position, or access to essential services. Qualified human review and appropriate professional advice remain necessary.
Systems and credentials
Only connect systems, accounts, data, and credentials you are authorized to use. Do not ask a Coworker to evade an approval, exceed its appointed role, conceal an external effect, or defeat a capability or policy boundary.
Enforcement
The effective policy will describe investigation, proportionate restriction or suspension, appeal, preservation, reporting, and emergency action. Serious or repeated violations may result in termination and, where legally required, reporting to the appropriate authority.
Questions about the product’s accountability model belong on the trust page.