Idiap and Logitech release DialFork benchmark for procedural hallucination

Idiap Research Institute and Logitech released DialFork, a benchmark for procedural hallucination in conversational agents: drifting from required steps even when tool calls are correct. It spans 24 tasks and 12 domains, with over 36,000 dialogues validated by 30 human annotators.

• Logitech CEO on AI agents and the great tech divide2026-10-10
• Treble raises $18M for voice simulation2026-09-17