Put AI on Your Side: AI-Powered Security on AWS
The same AI that hands attackers new capabilities is an even bigger lever for defenders. This piece frames security as threats meeting vulnerabilities, separates the three ways AI and security intersect, and shows four concrete AWS services for putting AI to work on defense.
The job has not changed, the ground under it has
Security is the work of keeping what we value safe. Strip away the tooling and the acronyms, and that is the whole job. The way I reason about it comes down to a simple working model: an incident happens when a threat meets a vulnerability.
Start with vulnerabilities, because they are the part we own. In NIST's terms, a vulnerability is a weakness in a system, its security procedures, internal controls, or implementation that could be exploited or triggered by a threat source. They fall into a few broad buckets: weak infrastructure configuration, vulnerable components inside the application, flawed application logic, and access misuse. That last one is a major problem in cloud security. Permissions are easy to grant and easy to forget, so accounts accumulate unused roles, stale credentials, and over-broad access that nobody actually needs. All of these buckets are controllable. The goal is straightforward to state and hard to do: ship to production with as few vulnerabilities as possible.
Threats are what come at those weaknesses. NIST defines a threat as any circumstance or event with the potential to adversely impact operations, assets, or individuals through an information system, whether by unauthorized access, destruction, disclosure, modification, or denial of service. In practice that is a long and familiar list: ransomware, distributed denial of service (DDoS) attacks, credential theft, malware, data exfiltration, cryptomining, and phishing, with new variants showing up constantly.
The asymmetry between the two is the reason security is hard. Threats are outside our control. We do not get to decide who attacks us, with what, or when, so our work against threats is mostly about detection, seeing them early, and where possible responding fast enough to block them. Vulnerabilities are the opposite. They are ours, and we can fix them before they ship.
For years, defending across both sides was a human-paced job. You configured, you scanned, you reviewed, you fixed, all at the speed people could work. AI broke that pace, on both sides of the fight.
AI splits security into three different conversations
When people say "AI and security" they usually mean one of three different things, and conflating them causes most of the confusion in the room.
The first is security of generative AI applications. When an organization builds a Gen AI product, it inherits a new class of threats and vulnerabilities that did not exist before, and it needs new controls to match. That trade is usually worth making, because the business value of the application justifies the added surface.
The second is security threats from AI. The same capability that increases business opportunity can also become a business risk. Adversaries use AI to write more convincing phishing, including synthetic voice, and to analyze and probe internet facing applications faster than before. AWS has documented this shift directly: today's internet is increasingly populated by automated AI bots, including AI agents that can autonomously navigate and interact with applications to perform complex tasks. Some of that traffic is useful. Some of it is reconnaissance.
The third is AI-powered security, using AI to strengthen our own operations across development, detection, and response. This is where I want to spend the rest of this story, for one reason. The defensive upside compounds. Used well, AI-powered security adds capability faster than AI adds risk, and that is what offsets the threat AI introduces in the first place.
The rest of this story shows four concrete ways to put that into practice on AWS.
Four places to put AI to work
Each of these maps to a control you likely already run. To make them easy to compare, I describe all four the same way: the control we used before AI, how the new service improves it, how the service actually works, and when to reach for it.
1. Detect threats with AI: Amazon GuardDuty
Before AI. Threats are the part we cannot control, so detection is a must. We cannot protect what we cannot see. The traditional control is to collect logs, then write rules and signatures to catch known-bad activity. That works only for the patterns you already thought to look for, and someone has to maintain those rules.
How the service helps. With Amazon GuardDuty, we detect threats continuously across accounts and workloads without building and maintaining that detection plumbing, and it catches anomalies that static rules would miss.
How it works. GuardDuty continuously monitors AWS accounts and workloads, consuming data sources like AWS CloudTrail event logs, VPC Flow Logs, and DNS logs. It combines threat intelligence feeds with machine learning models, learns the normal behavior of your environment, and surfaces the anomalies. It can detect compromised and exfiltrated credentials, unauthorized cryptomining, malware on EC2 instances and in container workloads, and unusual API call patterns that suggest an attacker is inventorying your resources. For these foundational data sources there are no agents to deploy or sensors to manage; deeper runtime visibility is available through an optional agent.
When to use it. Make it your baseline threat detection across every account. It is the first thing to turn on when you need broad coverage without standing up a detection engineering team.
2. Find vulnerabilities with autonomous penetration testing: AWS Security Agent
Before AI. The control for application vulnerabilities is scanning plus periodic manual penetration testing. Pentesting is slow, expensive, and scarce, so most organizations only test their most critical applications and leave the rest exposed between assessments.
How the service helps. With AWS Security Agent, we detect application vulnerabilities faster and far more often, on demand instead of once or twice a year.
How it works. AWS Security Agent, generally available since March 2026, is an autonomous penetration testing solution that deploys AI agents to discover, validate, and report vulnerabilities through multi-step attack scenarios. The key word is validate. Unlike a scanner that emits a pile of unconfirmed findings, it works like a human pentester: it identifies a potential weakness, then attempts to exploit it with targeted payloads and attack chains to confirm the risk is real. By ingesting your source code, architecture diagrams, and documentation, it understands how the application was built well enough to chain individual flaws into the higher severity attack paths that scanners miss. Each finding comes with a confidence rating and detailed reproduction steps, and AWS reports customers cutting testing timelines from weeks to hours.
When to use it. Reach for it when you release frequently and your manual pentest capacity only covers a handful of apps. Run it before pushing a new feature to production rather than waiting for the next scheduled assessment.
3. Block AI-driven attacks with an intelligent firewall: AWS WAF
Before AI. At the edge, the control is a web application firewall with static signatures and rate rules to block known-bad traffic. That struggles against bots that do not self-identify and against distributed, proxy-based attacks that rotate IP addresses to evade rate limits.
How the service helps. With AWS WAF intelligent threat mitigation, we block automated and AI-driven traffic that static rules let through.
How it works. The Bot Control managed rule group offers a targeted protection level that uses browser interrogation, fingerprinting, and behavior heuristics to spot bots that do not self-identify. A subset of these, the machine learning rules, analyze website traffic statistics like timestamps and browser characteristics to detect the distributed, coordinated bot activity behind proxy-based attacks, without you writing a single custom rule. AWS WAF has extended this into AI bot management and, as of February 2026, added an AI Activity Dashboard for visibility into AI bot traffic. The firewall is no longer matching static signatures. It is reasoning about behavior.
When to use it. Use it on internet-facing applications that face automated traffic: scraping, credential stuffing, account takeover, and account creation fraud. This is your real-time control at the edge while detection and remediation run behind it.
4. Remediate findings with AI assistance: Kiro
Before AI. The control is to detect misconfigurations with a posture tool, then remediate them by hand or with purpose-built automation like Lambda functions or Systems Manager runbooks. AWS Security Hub CSPM continuously checks your resources against security controls and reports failed configurations, but that backlog grows faster than humans can close it.
How the service helps. With Kiro, we remediate findings faster by handing the judgment-heavy work to an AI assistant while a human stays in control of the change.
How it works. When Security Hub CSPM raises a finding, such as a resource that violates a control, you hand it to Kiro and ask it to fix the underlying configuration. Kiro does the work of understanding the finding, locating the offending code or configuration, and proposing a correct change. You review and approve before anything is applied.
When to use it. Use it when your posture backlog outpaces your team, especially for findings that have a clear fix but still require code or configuration judgment to apply safely. The backlog stops being a wall and becomes a queue you can actually clear.
The takeaway: infusing AI into security is not optional anymore
Look at how software gets built now. Where an organization used to manage a handful of applications, developers vibe-code and launch many, and some of those are shadow applications nobody formally signed off on. The volume and velocity of what we have to defend has gone up sharply, and it is not coming back down.
This brings us back to the first conversation we opened with: building Gen AI capability. The teams moving fastest there have the most to gain from letting their security operations move at the same speed. When capability and defense advance together, the AI you ship stays an asset and the controls around it keep pace. That is the win to aim for.
You cannot meet that with headcount. The only thing that scales with AI-driven development is AI-driven defense. Detect threats with GuardDuty, find vulnerabilities with AWS Security Agent, block automated attacks with AWS WAF, and clear findings with Kiro. The same AI that hands attackers new capabilities hands defenders a larger one. Infusing AI into security operations is no longer a clever optimization. It is the natural, necessary step to keep our security in check.
References
- Amazon GuardDuty, threat detection overview and capabilities. AWS. https://aws.amazon.com/guardduty/ and https://docs.aws.amazon.com/guardduty/latest/ug/what-is-guardduty.html
- AWS Security Agent on-demand penetration testing is now generally available. AWS What's New, March 2026. https://aws.amazon.com/about-aws/whats-new/2026/03/aws-security-agent-ondemand-penetration/
- AWS Security Agent on-demand penetration testing now generally available. AWS Security Blog. https://aws.amazon.com/blogs/security/aws-security-agent-on-demand-penetration-testing-now-generally-available/
- AWS launches frontier agents for security testing and cloud operations. AWS Machine Learning Blog. https://aws.amazon.com/blogs/machine-learning/aws-launches-frontier-agents-for-security-testing-and-cloud-operations/
- AWS WAF intelligent threat mitigation. AWS DDoS resiliency best practices whitepaper. https://docs.aws.amazon.com/whitepapers/latest/aws-best-practices-ddos-resiliency/aws-waf-intelligent-threat-mitigation.html
- AWS WAF Bot Control rule group (Common, Targeted, and TGT_ML_ machine learning rules). AWS. https://docs.aws.amazon.com/waf/latest/developerguide/aws-managed-rule-groups-bot.html
- How to manage AI Bots with AWS WAF and enhance security (includes February 2026 AI Activity Dashboard update). AWS Networking & Content Delivery Blog. https://aws.amazon.com/blogs/networking-and-content-delivery/how-to-manage-ai-bots-with-aws-waf-and-enhance-security/
- SEC04-BP04 Initiate remediation for non-compliant resources, AWS Security Hub CSPM. AWS Well-Architected Framework. https://docs.aws.amazon.com/wellarchitected/latest/framework/sec_detect_investigate_events_noncompliant_resources.html
- Threat, definition. NIST Computer Security Resource Center Glossary. https://csrc.nist.gov/glossary/term/cyber_threat
- Vulnerability, definition (NIST SP 800-37 Rev. 2). NIST Computer Security Resource Center Glossary. https://csrc.nist.gov/glossary/term/vulnerability
This content publication is assisted by Claude Opus 4.8.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article