Skip to main content

Cloud Security

Rise of the blender: chaining low-severity vulnerabilities in the era of agentic AI

Agentic attackers will chain low- and medium-severity misconfigurations into full compromises. One worked example on AWS — nine findings, none critical, one breach — and what it means for prioritization.

Author
Ahsan MirCEO & Founder, Rapticore
Reviewed by
Rapticore Security EngineeringSecurity Engineering Reviewer
Published
Updated
Reviewed
Reading time
4 min

As we move into the world of agentic AI, I can imagine that we will witness a rapid rise in blended attacks — where attackers chain low- and medium-severity vulnerabilities to achieve full system compromises. This evolving threat landscape demands a fresh perspective on security posture and response strategies: rethink vulnerability prioritization.

Example of a blended attack: cloud misconfigurations leading to a compromise

Scenario: an attack on a SaaS company using AWS. A SaaS company running its infrastructure on AWS suffers a data breach through a blended attack that combines phishing, cloud misconfigurations, and privilege escalation to achieve full system compromise.

Initial access: a phishing attack on a DevOps engineer

  • The attacker sends a phishing email to a DevOps engineer, masquerading as an AWS IAM alert.
  • The email contains a malicious link leading to a fake AWS login page.
  • The engineer unknowingly enters their AWS IAM credentials.
  • The attacker gains valid AWS IAM user credentials — without MFA.

Best practice: enforce IAM MFA for all users. Misconfiguration (medium): the IAM user was not required to use MFA.

Privilege escalation: exploiting overly permissive IAM policies

  • The attacker logs into the AWS console using the stolen credentials.
  • They discover the compromised user has excessive IAM permissions, including the ability to create new IAM users, attach policies, and modify EC2 instances.
  • The attacker creates a new IAM user with AdministratorAccess and generates long-lived access keys for persistence.

Best practice: use least-privilege IAM policies and audit access regularly. Misconfiguration (high): the IAM user had wildcard permissions (iam:*).

Discovery: exploiting publicly exposed S3 buckets

  • The attacker uses the AWS CLI to list all S3 buckets (s3api list-buckets).
  • They discover a publicly exposed bucket containing database credentials in plaintext, API keys for third-party services, and environment variables for production workloads.

Best practice: enforce S3 bucket policies and restrict public access. Misconfiguration (high): an S3 bucket with public access enabled. Misconfiguration (low): secrets stored in plaintext instead of AWS Secrets Manager.

Lateral movement: exploiting open security groups

  • The attacker discovers an EC2 instance running a PostgreSQL database.
  • The security group allows open SSH (port 22) and database access (port 5432) from the internet.
  • Using the database credentials found in the bucket, the attacker logs into the database.
  • They dump sensitive customer data and create a reverse shell to gain remote access to the instance.

Best practice: restrict security group rules to specific IP ranges. Misconfiguration (high): SSH open to the world (0.0.0.0/0). Misconfiguration (medium): the database exposed to the internet.

Persistence: backdooring the environment

  • The attacker creates a new IAM role with full access to S3, EC2, and RDS.
  • They attach the role to the compromised EC2 instance to maintain persistence.
  • They also install a cryptocurrency miner on the instance to monetize the compromise.

Best practice: audit IAM roles and EC2 instance configurations regularly. Misconfiguration (medium): no monitoring or alerting for unusual IAM role creation.

Impact: data exfiltration and ransomware

  • The attacker exfiltrates sensitive customer data to an external S3 bucket under their control.
  • They encrypt production files on the EC2 instance and delete backups stored in S3 and EBS snapshots.
  • A ransom note is left demanding Bitcoin payment for data restoration.

Best practice: enable AWS CloudTrail logging, backup versioning, and Amazon GuardDuty. Misconfiguration (medium): CloudTrail logging disabled for S3. Misconfiguration (medium): no immutable backups or versioning enabled.

Attack path summary

The attack as a chain of six stages — initial access, privilege escalation, discovery, lateral movement, persistence, impact — each with the misconfiguration it rode on and its severity: one low, four medium, four high, no critical.
Six stages, nine misconfigurations, none of them critical. Each link is a finding a team would defer; together they were a breach.

Mitigation strategies

  • Enforce IAM MFA for all users.
  • Restrict IAM permissions using least-privilege access.
  • Disable long-lived IAM access keys; use temporary credentials with IAM roles.
  • Ensure all S3 buckets are private by default and use AWS Secrets Manager.
  • Lock down security groups and limit inbound access.
  • Enable AWS CloudTrail logging and monitor API calls.
  • Enforce data backup versioning and retention policies.
  • Deploy cloud detection and response for threat detection.

Prioritization is undoubtedly making a difference, but the future will focus on visibility and strong cyber hygiene — we need to be doing the basics. I would love to hear your thoughts.

Related content

  1. Cloud guardrails before detection: why prevention must happen at the boundary

    Cloud posture programs that stop at detection can spend weeks rediscovering the same classes of misconfiguration. As cloud and attacker velocity increase, the stronger model is to turn high-confidence security requirements into enforceable boundaries — preventing unsafe state where possible, detecting what cannot be prevented, and remediating the remainder under explicit policy.

  2. What you don’t know CAN hurt you

    A true incident story: an undocumented link from development to production, a zero-day, and sixteen hours in a war room. Why cloud risk starts with knowing what is actually in your environment.

  3. What’s in the cloud?

    No one knows what is in the cloud — especially your cloud. Why the industry chases the shiny new thing while the first control in every framework, asset management, stays unsolved.

  4. Democratize security

    Organizational silos give teams unequal access to data and decisions, and security finds out about the next billion-dollar bet two weeks before launch. The case for a shared, real-time view everyone can act on.