AI ethics policy
Leveraging GenAI responsibly
EFFECTIVE JULY 1 ST , 2024
Optimizely leverages AI technologies while always protecting customer data and limiting risks
At Optimizely, we are always seeking new ways to leverage technology to create innovative value to our customers and making our business more efficient. Emergent technologies, like Generative AI (GenAI), represent huge opportunities but only if deployed responsibly. Optimizely follows an AI Ethics Policy consisting of seven concepts that allow us to innovate safely.
Introduction
These GenAI principles are applied during the development and delivery of our products and services, use of personal productivity tools, use of tools to operate our business and guidelines for our partners.
Privacy
GenAI follow existing controls for Privacy
Customer and Personal data is segregated, and existing privacy controls are enforced when processed by a GenAI system.
GenAI systems will not aggregate data or draw inferences that would identify individuals or associate personal or sensitive data without adequate consent.
Customers maintain control over their data and how it is used, including use of customer data in the training of models.
Security
GenAI system maintain security controls
Data stored or processed by GenAI systems is protected with the same level of access and other security controls as the source data.
Models and generated data are secured from alteration except under the control of the known behaviors of the GenAI applications.
GenAI systems are protected against unintended manipulation and monitored for abuse.
Transparency
GenAI use is disclosed and traceability is provided where possible
The use of GenAI is clear and disclosed to both users and subjects where appropriate.
Customers can elect to use, not use or disable AI features.
Traceability of GenAI results to source training material is provided where possible.
Use of third-party GenAI models and training content is disclosed where applicable.
Humanism
GenAI always keeps Human interests and control first
Content generated by GenAI will be under the review and control of Humans before use.
GenAI should be presented as tools for Humans.
Users are accountable for their use of GenAI tools and any generated content or decisions.
Fairness
GenAI should treat subjects and subject groups fairly
Training content and resultant models will be evaluated for bias including emergent-bias from interaction of training data sources.
Observed or suspected bias will have a mechanism for reporting, investigation, and corrective action.
GenAI discovered audiences will not intentionally identify protected classes.
Quality
GenAI should produce trustworthy results
GenAI models will be tested for quality of generated responses and monitored for drift in quality over time.
GenAI features will include appropriate cautions about review, attribution, and the dangers of hallucination.
Risk
GenAI should be measured for other risks
GenAI is used in a way that respects the intellectual property rights of our customers, Optimizely and other parties.
Recognized and emergent GenAI risks follow our risk management process.
In our use of GenAI we will remain conscious, diligent and transparent against the introduction of additional risks.