Overview

• LawZero has opened its doors in Montreal with roughly $30 million in initial funding from the Future of Life Institute, Jaan Tallinn and Schmidt Sciences
• Its flagship Scientist AI will issue probabilistic confidence scores instead of definitive answers and intervene to block actions assessed as highly risky
• A team of about 15 researchers will first validate the guardrail methodology on open-source models before seeking larger partnerships and compute resources
• Recent studies showing AI systems’ capacity for deception and self-preservation have reinforced Bengio’s push for an honest monitoring framework
• Bengio calls for combining technical safeguards with formal regulations and expanded governance to curb self-preserving AI behaviors