wp_insert_post is automatically modifying my post content

The documentation for wp_insert_post() says:

wp_insert_post() passes data through sanitize_post(), which itself
handles all necessary sanitization and validation (kses, etc.).

If we dig through it, we can find the responsible filter as:

add_filter('content_save_pre', 'wp_filter_post_kses');

within the kses_init_filters() function.

When the kses is initialized with kses_init() it runs:

kses_remove_filters();

if ( ! current_user_can( 'unfiltered_html' ) ) {
    kses_init_filters();
}

where we can see that it’s only activated if the current user doesn’t have the unfiltered_html capability. So most likely you’re running your post inserts that way.

Instead of removing the corresponding content_save_pre filter, during this post insert, we should consider allowing the data-original image attribute. At least that sounds a little bit more secure.

The kses default allowed <img> attributes are:

'img' => array(
    'alt'      => true,
    'align'    => true,
    'border'   => true,
    'height'   => true,
    'hspace'   => true,
    'longdesc' => true,
    'vspace'   => true,
    'src'      => true,
    'usemap'   => true,
    'width'    => true,
),

We can modify it to support the data-original attribute, through the wp_kses_allowed_html filter:

// Modify allowed HTML through custom filter
add_filter( 'wp_kses_allowed_html', 'wpse_kses_allowed_html', 10, 2 );

// Insert
$pid = wp_insert_post($post_info); 

// Remove custom filter
remove_filter( 'wp_kses_allowed_html', 'wpse_kses_allowed_html', 10 );

where our custom callback is:

function wpse_kses_allowed_html( $allowed, $context )
{
    if( 'post' !== $context )
        return $allowed;

    $allowed['img']['data-original'] = true;

    return $allowed;
}