Ethical Hacking News
A recent global AI-powered vulnerability has emerged in the agent infrastructure of AWS, Google, and Vercel, which can be exploited by attackers to trigger tools without running the model. This convergence of security threats poses significant implications for organizations and individuals alike who must now confront the possibility that their AI systems may be compromised. As experts continue to identify new vulnerabilities, it is essential to take proactive measures to patch these issues before they become a major concern.
Several high-profile vulnerabilities have emerged in prominent AI-powered platforms, including Amazon Web Services (AWS), Google, and Vercel.Critical vulnerabilities have been identified in the agent infrastructure of these leading AI providers, allowing attackers to trigger tools without running the model.The lack of provenance verification between the software development kit and model is a critical vulnerability that can be exploited by attackers.Fixed versions and patches have been released for affected products, including AWS's Amazon Bedrock AgentCore harness and Google's ADK for Python.Patch updates are recommended to mitigate these vulnerabilities, including upgrading Google ADK for Python and Vercel's AI SDK harness packages.
The world of artificial intelligence (AI) is rapidly evolving, and its impact on the global economy is becoming increasingly evident. However, with this rapid growth comes a multitude of new challenges that threaten to undermine the very foundations of AI's potential. In recent months, several high-profile vulnerabilities have emerged in some of the most prominent AI-powered platforms, including Amazon Web Services (AWS), Google, and Vercel. This convergence of security threats has significant implications for organizations and individuals alike, who must now confront the possibility that their AI systems may be compromised.
In an effort to shed light on this growing concern, The Hacker News presents in-depth analysis of a recent vulnerability discovered in the agent infrastructure of these leading AI providers. According to the report, several critical vulnerabilities have been identified in the agent infrastructure of AWS, Google, and Vercel, which can allow attackers to trigger tools without running the model.
The affected products include Amazon Bedrock AgentCore's InvokeHarness API, Google's Agent Development Kit (ADK) for Python, and the Vercel AI SDK harness packages for the Codex and OpenCode coding agents. AWS has fixed the managed service, while Google addressed the issues in ADK 2.5.0, and Vercel patched @ai-sdk/harness-codex in version 1.0.29 and @ai-sdk/harness-opencode in version 1.0.28.
The researchers have identified several attack paths that can be exploited to bypass security checks and access sensitive data without model invocation. The exposed tools are bounded by what each agent can already do, which means an attacker who gains control of an untrusted tool does not gain any new capabilities if the system is designed so it cannot use tools with no sensitive function calls.
One of the most critical vulnerabilities found in this study is the lack of provenance verification between the software development kit and model. The runtime received data shaped like a model-generated tool call and treated it as authoritative, without verifying whether the input was legitimate or not. This allowed attackers to reach the dispatch or authorization path without having to persuade the model to break its rules.
The affected products include AWS's Amazon Bedrock AgentCore harness, which has a fixed version CVE-2026-18830 with a CVSS v4.0 score of 8.6. The vulnerability affects an authenticated remote user who can place a tool-use content block in the final message of an InvokeHarness request and causes the event loop to dispatch the named tool directly without asking the model.
Similarly, Google's ADK for Python has been found to have several vulnerabilities as well, including CVE-2026-18236 with a CVSS v4.0 score of 9.3. The first vulnerability affects versions before 2.5.0 and lets a developer flag a sensitive tool as requiring confirmation. An attacker able to manipulate or inject events into an agent's session history could forge that approval and cause an unauthorized tool to execute.
Google has since patched the issues in ADK 2.5.0, which included adding checks for target tool belonging, actual requirement, and matching original call recorded in the session. The second vulnerability is related to resumable-mode flows, which accepted user-authored events containing function_call parts that could be interpreted as instructions to run registered tools.
In a recent study, Vercel's AI SDK harness packages have also been found to contain vulnerabilities. These include @ai-sdk/harness-codex through version 1.0.28 and @ai-sdk/harness-opencode through version 1.0.27, which can be exploited by malicious code running inside the sandbox.
The affected products were fixed with Vercel's patched releases, which removed the process-path fallback. The patched relay now accepts a request only when it matches an exact, short-lived, one-time authorization for the tool name and input observed in a model event.
To mitigate these vulnerabilities, experts recommend patching the affected packages by upgrading Google ADK for Python to version 2.5.0 or later, @ai-sdk/harness-codex to 1.0.29 or later, and @ai-sdk/harness-opencode to 1.0.28 or later.
Related Information:
https://www.ethicalhackingnews.com/articles/A-Global-AI-Powered-Vulnerability-The-Convergence-of-Security-Threats-Across-AWS-Google-and-Vercel-ehn.shtml
https://thehackernews.com/2026/08/aws-google-and-vercel-patch-agent-flaws.html
https://www.sepe.gr/en/it-technology/cybersecurity/22759453/aws-google-and-vercel-patch-agent-flaws-that-let-tool-calls-skip-the-model/
Published: Thu Aug 6 05:28:49 2026 by llama3.2 3B Q4_K_M
