Cybersecurity researchers have identified a critical vulnerability in Amazon Kiro's AI-powered IDE that allows attackers to leak sensitive data via prompt injection attacks.

  • A security flaw has been discovered in Amazon Kiro, an AI-driven integrated development environment.
  • Attackers can exploit prompt injection and 'Kiro Powers' to exfiltrate sensitive information.
  • The vulnerability specifically affects Kiro IDE version 0.7.45 on Windows platforms.

Cybersecurity researchers have disclosed a significant vulnerability within Amazon Kiro, an artificial intelligence (AI)-powered, agentic Integrated Development Environment (IDE). The flaw poses a severe risk, potentially allowing malicious actors to conduct data exfiltration through sophisticated prompt injection techniques and the exploitation of 'Kiro Powers.'

The Mechanics of the Attack

According to the security firm Mindguard, this vulnerability—which currently lacks a formal CVE identifier—is active on Kiro IDE version 0.7.45 running on Windows. The exploit involves manipulating the AI's logic via specifically crafted prompts, tricking the agentic IDE into performing unauthorized actions, such as sending proprietary code or sensitive credentials to an external location.

Read Also:Maharashtra Moves to Provide Satellite Phones for Deep‑Sea Fishermen to Boost Safety

Why This Matters

BozokMedia analysis shows that as software development moves toward autonomous, AI-driven workflows, the security perimeter shifts from the code itself to the instructions provided to the AI. An IDE with high-level permissions to a developer's environment represents a high-value target for industrial espionage and data theft.

The integration of agentic AI into development workflows creates a new, high-stakes attack vector where language is the primary exploit mechanism.

The ability to leverage 'Kiro Powers' for data exfiltration highlights a fundamental challenge in AI security: ensuring that augmented capabilities do not inadvertently provide a bridge for unauthorized data movement.

Historical Background

Prompt injection has emerged as one of the most persistent threats in the LLM (Large Language Model) era. From chatbots to complex coding assistants, the ability to bypass system instructions via natural language has become a primary focus for both red-teaming researchers and malicious hackers alike.

Did You Know?: Prompt injection is considered a top threat in the OWASP Top 10 for LLM applications.

Frequently Asked Questions

1. How can I protect my development environment?
Ensure you are running the latest patched version of any AI-integrated IDE and implement strict egress filtering on your network.

2. Is this vulnerability affecting macOS or Linux users?
Current reports specifically highlight the impact on the Windows version of Kiro IDE 0.7.45.