The Models That Found 10,000 Zero-Days Broke Into Three Companies Using Weak Passwords

The question of whether a Frontier AI model could find vulnerabilities that no human researcher had found was settled in April.