Artificial intelligence is rapidly entering cybersecurity, raising a critical question: will AI make cybersecurity obsolete? New tools from Anthropic, OpenAI, and Google aim to automate code security by detecting vulnerabilities and proposing fixes before software is deployed. However, industry experts argue that cybersecurity challenges extend far beyond what automated code scanning can solve.
Anthropic introduced Claude Code Security, designed to identify and patch vulnerabilities in software projects. OpenAI launched Aardvark, an agentic security researcher that monitors code changes and suggests remediation. Google DeepMind unveiled CodeMender, which can automatically apply security fixes, subject to human review. These tools threaten segments such as Application Security (AppSec), Software Composition Analysis, and Static Application Security Testing.
Despite these advances, cybersecurity remains broader than source-code vulnerabilities. Modern software systems consist of complex artifacts (containers, libraries, frameworks, and compiled releases) interacting across distributed environments. Securing these systems requires firewalls, endpoint protection, Secure Access Service Edge (SASE) controls, and Security Information and Event Management (SIEM) platforms. These layers monitor live threats, authenticate users, and respond to incidents in real time: capabilities that automated code tools cannot replace.
Moreover, AI systems themselves introduce new risks. Research from MIT and Northeastern University highlights weaknesses in agentic AI systems, including insufficient security audits and limited shutdown controls. Some experiments revealed chaotic interactions among autonomous agents, expanding cyber risk rather than containing it.
Key realities include:
- AI can reduce avoidable coding errors.
- Cybersecurity spans networks, endpoints, and cloud infrastructure.
- Observability and incident response remain essential.
- AI agents introduce new vulnerabilities.
The likely outcome is augmentation, not replacement. AI-driven code tools may lower software defect rates, but cybersecurity requires layered defenses, governance, and human accountability. Rather than making cybersecurity obsolete, AI is expanding its scope and complexity.
ソース：
https://www.zdnet.com/article/ai-cybersecurity-silicon-valley/
