Google publishes its AI principles

Background

As AI moved from research labs into products used by large numbers of people, companies that build it came under pressure to say publicly what they would and would not do with it. Google's answer was a blog post titled "AI at Google: our principles", published on 7 June 2018 and written by CEO Sundar Pichai^[1].

What the post set out

The post lists seven objectives for AI applications. Google says these are meant to actively guide its research and product development and to shape business decisions, rather than serve as a general statement of values^[1]. The seven are:

• be socially beneficial;
• avoid creating or reinforcing unfair bias;
• be built and tested for safety;
• be accountable to people;
• incorporate privacy design principles;
• uphold high standards of scientific excellence;
• be made available for uses that accord with these principles^[1].

Some of the objectives come with a stated test. On social benefit, Google says it will proceed only where the overall likely benefits substantially exceed the foreseeable risks and downsides^[1]. On bias, it says it will seek to avoid unjust impacts on people based on sensitive characteristics such as race, ethnicity, gender, nationality, income, sexual orientation, ability, and political or religious belief^[1].

The uses Google ruled out

The part of the post that attracted the most attention was a list of four kinds of application Google said it would not pursue: technologies that cause or are likely to cause overall harm; weapons or other technologies whose principal purpose is to cause or directly facilitate injury to people; technologies that gather or use information for surveillance violating internationally accepted norms; and technologies whose purpose contravenes widely accepted principles of international law and human rights^[1].

The post also draws a boundary around that list. It clarifies that Google will still work with governments and the military in areas such as cybersecurity, training, veteran healthcare, and search and rescue^[1]. The excluded categories are defined by purpose and effect, not by the identity of the customer.

Why it mattered

A public list of this kind does several things at once. It tells employees and outside researchers what standard to hold the company to, it tells customers and governments what to expect, and it gives the company a written basis for declining work. Because the commitments are phrased as objectives to be weighed rather than as fixed rules, they leave room for judgement: the social-benefit test, for example, asks for a comparison of likely benefits against foreseeable risks rather than naming prohibited products.

The document was not frozen in place. The page carries a later update notice recording that the principles were revised on 4 February 2025^[1], which shows that the framework has been treated as something to be revisited as the technology and its uses changed.

What it is not

The principles are a statement of intent by one company, not a regulation or a technical standard, and the post describes them as guidance for Google's own work^[1]. For readers trying to understand how AI companies govern themselves, the document is a useful primary source: it shows which concerns a major developer chose to name in 2018 — benefit versus risk, bias, safety, accountability, privacy, scientific quality, and the limits of acceptable use — and it shows the wording it chose for each.