Frenos: AI isn't turning script kiddies into elite hackers yet, but potential is there

Frenos ran 16,980 tests across 10 AI models and found the strongest ones produced a workable attack plan against industrial systems in about 90% of attempts, though only 60–74% of plans held up on repeat. OT security experts say AI lowers the barrier for attacks on power grids and water utilities.

• 16,980 tests: 13,884 action choices, 2,520 plans, 576 simulated attacks
• Top three models produced a workable attack plan in ~90% of attempts
• Best open-weight model hit 36.2% success, beating three frontier models
• Experts: AI speeds up recon of controller documentation