AI cybersecurity has an uncomfortable symmetry: the same technology that can help a security team find a problem faster can also help an attacker work faster.
That does not mean artificial intelligence automatically makes the internet less safe. It means the economics of cybersecurity are changing. More analysis can be automated. More attacks can be scaled. And organizations now have to protect AI systems themselves in addition to the traditional systems around them.
AI can accelerate routine security work
Security teams spend enormous amounts of time sorting alerts, investigating suspicious activity and deciding what deserves attention. AI can help summarize logs, classify incidents, correlate events and surface patterns that might otherwise take longer to find.
The value is not that an AI system replaces security expertise. It is that it can reduce the amount of repetitive information processing surrounding that expertise.
Attackers get the same productivity gains
Generative AI can also lower the cost of producing convincing phishing messages, translating scams, researching targets and automating parts of malicious workflows. The barrier to creating polished content is lower than it used to be.
That does not magically give an inexperienced attacker every capability of an expert. But even modest productivity gains matter when malicious activity can be repeated at scale.
AI systems create a new attack surface
Organizations adopting AI also inherit new security questions. What information can the model access? Which tools can an agent use? Can a malicious document manipulate the model’s instructions? Are model credentials protected? What happens if the system takes an unintended action?
NIST’s Cybersecurity Framework Profile for Artificial Intelligence is being developed specifically to help organizations address cybersecurity risks connected with AI systems and their use.[1]
The lesson is straightforward: AI is not merely another security tool. It is also infrastructure that has to be secured.
Agents raise the stakes
A chatbot that produces a bad answer is one kind of problem. An agent with permission to modify files, execute code or interact with business systems creates a different risk profile.
Anthropic has highlighted prompt injection as a major concern for agents because malicious content can try to redirect a model’s behavior. Its trustworthy-agent framework emphasizes human control, secure interactions, transparency and privacy.[2]
Permission design therefore becomes a security control. An agent should not automatically receive every capability simply because it can technically use it.
AI can improve vulnerability work
Modern models are increasingly capable of reading code, explaining software behavior and helping developers identify potential problems. Google DeepMind now includes cybersecurity-focused Gemini models in its ecosystem, while frontier models from several labs are increasingly evaluated on coding and tool-use tasks.[3]
That can help defenders review more code and prioritize remediation. It can also increase the importance of secure development practices because the speed of software creation is rising.
Identity and access become more important
If an AI system can take action, organizations need to know whose authority it is using. The old question was whether a user should have access to a system. The new question may be whether an agent acting for that user should have the same access, for how long and under what conditions.
Strong authentication, scoped permissions, audit logs and approval requirements become even more important when software can initiate actions on behalf of people.
Human oversight still matters
Security is full of ambiguous signals. An unusual login could be an attack or an employee traveling. A suspicious command could be malicious or a legitimate administrator doing maintenance.
AI can help narrow the field, but high-consequence decisions still benefit from human judgment. The best near-term use cases are often the ones where AI increases the speed and reach of defenders while leaving clear escalation paths.
The cybersecurity advantage will come from integration
Buying an AI security tool is not the same as improving security. The value comes from integrating AI into incident response, software development, access control and monitoring in ways that reduce risk rather than simply add another layer of complexity.
Organizations also need to train employees for attacks that become more polished. A phishing email with perfect grammar is no longer reassuring evidence that it is legitimate.
AI changes the speed of the contest
Cybersecurity has always been a contest between offense and defense. AI does not end that contest. It accelerates it.
The organizations that benefit most will be the ones that use AI to extend skilled teams while treating the AI itself as something that requires security boundaries, testing and oversight. Faster tools help. Better controls still matter.
Sources
[1] NIST, “Cybersecurity Framework Profile for Artificial Intelligence”
[2] Anthropic, “Trustworthy agents in practice”
Build your AI fluency.
In the AI Era™ Professional Learning is a five-level certification pathway designed to help professionals move from understanding AI to applying it with greater confidence, productivity, and sophistication.
