tech
The next phase of AI cybersecurity still needs humans
The new phase of AI-powered cybersecurity depends on how effectively humans can direct these models.
TL;DR
- AI models like Anthropic's Mythos and OpenAI's GPT-5.5-Cyber are powerful tools for finding software bugs and writing exploits.
- Companies testing these models found a significantly higher number of bugs compared to their usual rates.
- These AI systems are effective at linking low-severity vulnerabilities into workable attack chains.
- Despite their power, the AI models require human expertise to validate findings, guide workflows, and assess the practical significance of vulnerabilities.
- A false positive rate of around 30% was observed, which decreased as models were trained on specific environments.
- Cisco's blueprint for AI security suggests instructing models to make 'checkable' claims and verify their own findings to manage unreliable output.
- While attackers may not face the same learning curve, human oversight remains crucial for effective and reliable use of AI in cybersecurity.