AWS Strands Agents Tools Exposed to Multiple CVEs in 23 Days

• •AWS Strands Agents Tools had four CVEs published between July 15 and August 6, 2026.
• •The vulnerabilities expose security-sensitive parameters, enabling credential exfiltration and command execution.
• •Remediations have been applied, but the root cause indicates a broader design flaw in the system.

Between July 15 and August 6, 2026, AWS Strands Agents Tools received four CVEs due to a design flaw exposing security-sensitive parameters as LLM-controllable inputs. The vulnerabilities include CVE-2026-15746 (credential exfiltration), CVE-2026-18394 (proxy hijacking), CVE-2026-18733 (arbitrary command execution), and CVE-2026-19111 (tenant-memory forgery). These flaws affect tools like elasticsearch_memory and http_request, allowing attackers to manipulate parameters and execute commands or access unauthorized data. The severity of these vulnerabilities ranges from 6.5 to 8.8 on the CVSS scale, indicating significant risks to cloud, fintech, and crypto/DeFi infrastructures. Remediations have been implemented, but the systemic nature of the flaws raises concerns about agent identity governance. AWS has classified these issues under CWE-1427, highlighting the risks associated with LLM-integrated systems.