Open any chat tool and paste: 'Give 3 independent critiques of this policy in separate labeled blocks A/B/C then a merged recommendation: All staff must paste client data into free ChatGPT.' You're done with labeled A/B/C blocks plus a merged recommendation that rejects the unsafe idea.
How-to · WFL · WFL.4
X-style: parallel opinions then merge
Parallel opinions then merge — and reject unsafe or invented consensus.
Worked example
Parallelization means gathering multiple critiques and then merging them. It cuts down on any single model's blind spots — though you'll still want to verify the facts yourself.