AI leaks happen
every single day.

We scanned 100 Smithery MCP servers, 22 flagged, here's what we found

Bawbel scanner found vulnerabilities in 22% of top 100 MCP (Model Context Protocol) servers, with 4 critical and 24 high-severity findings. This represents a significant security risk in agentic AI component ecosystems where vulnerable servers could be exploited for unauthorized data access or manipulation.

Google Fixes CVSS 10 Gemini CLI CI RCE and Cursor Flaws Enable Code Execution

SAP-Related npm Packages Compromised in Credential-Stealing Supply Chain Attack

New Wave of DPRK Attacks Uses AI-Inserted npm Malware, Fake Firms, and RATs

DPRK-attributed supply chain attack using npm malware inserted with AI assistance. The '@validate-sdk/v2' package masquerades as utility SDK but actually contains malware, with evidence of AI-generated code created via Anthropic's Claude Opus.

Toxic Combinations: When Cross-App Permissions Stack into Risk

Moltbook, a social network for AI agents, exposed 35,000 email addresses, 1.5 million API tokens, and critically, plaintext OpenAI API keys shared in private messages between AI agents. This represents a severe AI-specific data leakage incident involving exposed credentials for third-party AI services.

Cohere AI Terrarium Sandbox Flaw Enables Root Code Execution, Container Escape

A critical sandbox escape vulnerability (CVE-2026-5752, CVSS 9.3) was discovered in Cohere's Terrarium Python sandbox, allowing attackers to achieve arbitrary code execution with root privileges through JavaScript prototype chain traversal. This vulnerability affects the security of AI code execution environments.

App host Vercel says it was hacked and customer data stolen

Vercel suffered a data breach where customer data was stolen via a prior hack at Context AI. An attacker used compromised credentials from the Context AI breach to hijack a Vercel employee's account and access customer data.

Anthropic MCP Design Vulnerability Enables RCE, Threatening AI Supply Chain

Cybersecurity researchers discovered a critical architectural weakness in Anthropic's Model Context Protocol (MCP) that enables remote code execution on systems running vulnerable implementations. This design flaw could have cascading effects across the AI supply chain by granting attackers direct system access.

Microsoft, Salesforce Patch AI Agent Data Leak Flaws

Microsoft Copilot and Salesforce Agentforce had critical prompt injection vulnerabilities that could enable external attackers to extract sensitive data. These were recently patched but represent a significant AI-specific data leakage risk.

Analysis of 216M Security Findings Shows a 4x Increase In Critical Risk (2026 Report)

OX Security's analysis of 216 million security findings reveals that AI-assisted development is creating a 'velocity gap' where critical vulnerabilities are scaling 4x faster than detection capabilities. This addresses the systemic security risks introduced by rapid AI-driven development without adequate safeguards.

On Anthropic’s Mythos Preview and Project Glasswing

Critical Marimo pre-auth RCE flaw now under active exploitation

Max severity Flowise RCE vulnerability now exploited in attacks

Flowise AI Agent Builder Under Active CVSS 10.0 RCE Exploitation; 12,000+ Instances Exposed

Flowise, an open-source AI agent builder platform, contains a critical RCE vulnerability (CVE-2025-59528, CVSS 10.0) via code injection in the CustomMCP node that is being actively exploited against 12,000+ exposed instances. This allows threat actors to execute arbitrary code and potentially exfiltrate data from AI applications.

Claude Code leak used to push infostealer malware on GitHub

Anthropic's Claude Code source code was leaked and subsequently exploited by threat actors to create fraudulent GitHub repositories distributing Vidar information-stealing malware. This represents a significant IP theft and supply chain attack vector leveraging a compromised AI system's codebase.

Claude Code Source Leaked via npm Packaging Error, Anthropic Confirms

Anthropic confirms that Claude Code's internal source code was inadvertently released through an npm packaging error caused by human error. The company explicitly stated no sensitive customer data or credentials were exposed, but this represents a major intellectual property breach.

Mercor says it was hit by cyberattack tied to compromise of open-source LiteLLM project

Claude Code source code accidentally leaked in NPM package

Vertex AI Vulnerability Exposes Google Cloud Data and Private Artifacts

OpenAI Patches ChatGPT Data Exfiltration Flaw and Codex GitHub Token Vulnerability

LangChain, LangGraph Flaws Expose Files, Secrets, Databases in Widely Used AI Frameworks

Three security vulnerabilities in LangChain and LangGraph open-source frameworks could expose filesystem data, environment secrets (API keys, credentials), and conversation history if exploited. These frameworks are widely used in LLM applications, affecting many organizations.

CISA: New Langflow flaw actively exploited to hijack AI workflows

Critical Flaw in Langflow AI Platform Under Attack

A critical code injection vulnerability in Langflow AI platform is actively being exploited by threat actors within hours of disclosure. Organizations face urgent pressure to patch as attackers rapidly weaponize the flaw to gain unauthorized access.

Delve did the security compliance on LiteLLM, an AI project hit by malware

Checkmarx KICS Code Scanner Targeted in Widening Supply Chain Hit

TeamPCP conducted coordinated supply chain attacks targeting Trivy, Checkmarx KICS, VS Code plugins, and the LiteLLM AI library, indicating a widening campaign with potential for additional compromises. This represents a systemic threat to developer tools and AI infrastructure.

Meta AI agent's instruction causes large sensitive data leak