Social media companies are facing intense scrutiny over their use by foreign agents to influence U.S. politics. That includes pending legislation to force them to disclose the sources of political advertising, and broad pressure to identify and curtail users who aren’t who they claim to be.
Now those issues appear to have come home to roost in dramatic fashion. The Daily Beast reported Friday that in March of 2016, Twitter CEO Jack Dorsey retweeted an account twice that has been identified by Russia’s RBC news outlet as a “troll” created by Russia’s Internet Research Agency – also known as its “troll farm.”
The account in question, @Crystal1Johnson, seemed at first glance to belong to an African-American woman, and Dorsey retweeted two of its more upbeat messages – one about racial tolerance, and one celebrating the singer Rihanna.
The account has now been suspended, supporting the idea that it wasn’t quite what it claimed to be. But archived tweets unearthed by the Daily Beast included much more extreme messages, including one from May 2016 connecting then-Presidential candidate Hillary Clinton to the Ku Klux Klan.
Get Data Sheet, Fortune’s technology newsletter.
The incident highlights the difficulty of distinguishing between grassroots political activism and foreign meddling on social media. Many of @Crystal1Johnson’s tweets seem to echo the Black Lives Matter movement’s campaign against police abuse of African Americans. But others, including a tweet propagating the false idea that William Shakespeare’s plays were secretly written by a black woman, could just as easily be seen as part of a campaign to heighten racial discord.
The depth and breadth of Russia’s social media propaganda is still being uncovered. The Kremlin was also recently found to have impersonated or subverted rappers and American Muslims to sow rancor and spread misinformation about U.S. politicians, particularly Hillary Clinton.
Pressure is ramping up to hold companies like Twitter accountable for such misuse of their platforms – but Dorsey’s retweets suggest even they might have a hard time telling frauds from real users.