Jansi already falls back to a stream that strips the ANSI sequences if WindowsAnsiOutputStream fails to initialize - I'm unclear on why Logback attempts to initialize it directly. Here's a custom ConsoleAppender I'm using to avoid issues with the symptoms raised in this bug in environments where WindowsAnsiOutputStream is not supported:

import java.io.OutputStream;
import org.fusesource.jansi.AnsiConsole;
import ch.qos.logback.core.ConsoleAppender;

public class AnsiConsoleAppender<E> extends ConsoleAppender<E> {

	@Override
	public void setOutputStream(OutputStream outputStream) {
		super.setOutputStream(AnsiConsole.wrapOutputStream(outputStream));
	}
}