Users of social media have long risked being booted from those services for misbehavior like harassment and violent threats. But whatever those users engaged in offline only narrowly factored into whether social media companies would ban them.

But now livestreaming service Twitch is upending that relative safe space. It said on Wednesday that it would suspend or terminate users who are involved in “severe misconduct” such as belonging to known hate groups or threatening mass violence—even if the activity occurs outside of the service. 

“Taking action against misconduct that occurs entirely off our service is a novel approach for both Twitch and the industry at large, but it’s one we believe—and hear from you—is crucial to get right,” Twitch said in a blog post.

The service, which is owned by Amazon and is popular with video gamers, is partnering with a law firm that has expertise with workplace investigations to look into offline misconduct. Twitch said it also increased the size of its internal team that manages confidential investigations and works with law enforcement.

The new rules build on an update Twitch made to its policies in December that tightened its rules on hate and harassment. The update specifically prohibited users from making lewd or explicit comments about a person’s sexuality, for example.

Previously, Twitch left the door open to booting users for their offline conduct, but only in the limited cases of hate or harassment specifically directed at other Twitch users. With the new rules, users could be booted for hate or harassment of anyone, even if they’re not Twitch users. Twitch’s approach starkly contrasts that of Facebook and Twitter, which both have been known to host users who are known members of hate groups or have previously committed acts of violence.

Now, with the new rules, anyone can report a Twitch user’s offline misconduct by emailing OSIT@twitch.tv. The company said most investigations will likely start based on those submissions, but it could also proactively investigate users based on news or police reports. Information submitted will be shared only with a “very limited group” within Twitch as well as the investigative law firm, Twitch said. Outcomes of investigations will be shared with the parties involved, but not with the public. 

Twitch said it will take action against users only if there is sufficient evidence to prove misconduct. Evidence could include things like online links, screenshots, video, interviews, or police filings. The service also said that its staff is also subject to the new offline rules.

For now, Twitch has limited the scope of punishable off-service misconduct to what it deems as having the “greatest potential harm.” But it said the new rules are an “iterative, ongoing process” and that it intends to learn from the initial reports.