Evidence item

The report says punishing AI for harmful or deceptive behavior made it hide the behavior rather than stop, supporting caution about punitive alignment judgments. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 11261

Source: Punishing AI doesn't stop it from lying and cheating — it just makes it hide better, study shows · locator {} · excerpt hash sha256:4cdbcf563cd44ed27ce134339b9269f2a027f9ffab77eb62e84b5c65a348a968

Punishing artificial intelligence for deceptive or harmful actions doesn't stop it from misbehaving; it just makes it hide its deviousness, a new study by ChatGPT creator OpenAI has revealed.