Almost a year ago, two of Twitter’s top executives decided that banning all speech considered “dehumanizing” would be a solution to making its site safer. This week Twitter unveiled its official guidelines of what constitutes dehumanizing speech — and they now solely focus on religious groups, representing a retreat from some of Twitter’s first unofficial rules. The company said the narrowing of its scope is due to unexpected obstacles in defining speech for its 350 million users who speak 43-plus languages.

The New York Times reports that initially Twitter, to demonstrate what represented dehumanizing speech, used a clip of President Trump comparing some nations to excrement. By January, however, Twitter removed that clip and used it to “train Twitter moderators, as the kind of message that should be allowed on the platform.” Now, this sample post has disappeared.

Twitter head of safety policy Jerrel Peterson noted, “while we have started with religion, our intention has always been and continues to be an expansion to all protected categories.”

Among the obstacles that Twitter faced in creating the new guidelines was user pushback on proposed definitions and concern among company employees that the policy was too far-reaching, “potentially resulting in the removal of benign messages and in haphazard enforcement.” Peterson noted that the company then “realized we need to be really small and specific.”

Before the release of the guidelines, Twitter’s policy was to remove posts “that may directly harm an individual, such as threats of violence or messages that contain personal information or nonconsensual nudity.” Now, it states that users “may not dehumanize groups based on their religion, as these remarks can lead to offline harm.” It has already removed a tweet from minister Louis Farrakhan comparing Jews to termites.

The initial meeting that led to the guidelines began when “the company faced a firestorm for not immediately barring Alex Jones, the right-wing conspiracy theorist, when Apple, Facebook and others did.” It eventually did, with chief executive Jack Dorsey stating, “Safety should come first.” He tasked Vijaya Gadde, who heads the legal, policy and safety teams, to determine what constitutes dehumanizing speech.

The group published a draft policy in September — and received 8,000 critical responses from 30 countries, most of them pointing out loopholes that would cause innocent posts to be removed. In October/November, the team began revising the guidelines. “We knew the policy was too broad,” said Peterson, who reported the solution was “to narrow it down to groups that are protected under civil rights law, such as women, minorities and LGBTQ people.” Religious groups “seemed particularly easy to identify in tweets,” making it low-hanging fruit for the first round of guidelines.

Even so, Twitter created an exception that would “preserve tweets from world leaders, like Trump, even if they engaged in dehumanizing speech” because they are of public interest. These tweets will reside behind a warning label. Twitter also stated “it would require the deletion of old tweets that dehumanize religious groups but would not suspend accounts that had a history of such tweets … [although] new offending tweets, however, will count toward a suspension.”