Strange UTF-8 processing changes when updating to Oracle Java 8

Question

Function being tested:

public static String removeNonprintableCharacters(String input) {
    StringBuilder newString = new StringBuilder(input.length());
    for (int offset = 0; offset < input.length();) {
        int codePoint = input.codePointAt(offset);
        offset += Character.charCount(codePoint);

        // Replace invisible control characters and unused code points
        switch (Character.getType(codePoint)) {
            case Character.CONTROL:     // \p{Cc}
            case Character.FORMAT:      // \p{Cf}
            case Character.PRIVATE_USE: // \p{Co}
            case Character.SURROGATE:   // \p{Cs}
            case Character.UNASSIGNED:  // \p{Cn}
                newString.append("\ufffd");
                break;
            default:
                newString.append(Character.toChars(codePoint));
                break;
        }
    }
    return newString.toString();
}

Test method:

@Test
public void testRemoveNonprintableCharacters() throws UnsupportedEncodingException {
    assertEquals("\ufffd", r(new byte[]{0}));
    // jdk7:
    //assertEquals("\ufffd", r(new byte[]{-7, 'a'}));
    // jdk8: (???)
    assertEquals("\ufffda", r(new byte[]{-7, 'a'}));
}

private String r(byte[] bytes) throws UnsupportedEncodingException {
    return Unicode.removeNonprintableCharacters(new String(bytes, "UTF-8"));
}

As you can see in the test method, the returned result is different after upgrading the JVM to Java 8... why?