Posts

Capsule Security — ShareLeak and PipeLeak prompt injection in Copilot Studio and Agentforce

Capsule Security disclosed CVE-2026-21520 (ShareLeak) in Microsoft Copilot Studio and PipeLeak in Salesforce Agentforce — indirect prompt injection via public forms leading to data exfiltration.

OpenClaw — CVE-2026-33579 Privilege Escalation via Missing Scope Validation in Device Pairing

ChatboxAI — MCP StdioClientTransport OS Command Injection (CVE-2026-6130)

OpenAI — Axios Supply Chain Compromise Impacts macOS App Certification

OpenAI discloses Axios npm supply chain attack affecting GitHub Actions workflow with access to macOS app signing certificates, highlighting critical AI infrastructure supply chain risks.

arXiv — MCP-DPT: Defense Placement Taxonomy for Model Context Protocol Security

New arXiv research introduces MCP-DPT, a defense-placement taxonomy that maps 49 MCP-specific attacks to architectural layers and identifies critical security gaps in AI agent ecosystems.

GitHub Advisories — MCP Security Vulnerabilities Cluster (CVE-2026-40159, CVE-2026-39885)

GitHub Security Advisories disclose two critical MCP vulnerabilities: PraisonAI environment exposure and FrontMCP SSRF, highlighting systemic risks in AI agent toolchains.

Anthropic — Command injection vulnerability fixed in Claude Code LSP binary detection

Trend Micro — Sockpuppeting: Single-Line Jailbreak for 11 Major AI Models

Trend Micro researchers discovered sockpuppeting — a single-line code jailbreak that bypasses safety guardrails in 11 LLMs including ChatGPT, Claude, and Gemini with up to 15.7% success rate.

Google Chrome — Critical WebML and PrivateAI vulnerabilities expose memory data and enable sandbox escape

Two critical Chrome vulnerabilities (CVE-2026-5885 and CVE-2026-5874) in AI components expose memory data and enable sandbox escape attacks, highlighting security risks in browser-based machine learning.

Guardian — Claude Mythos AI model demonstrates unprecedented vulnerability discovery capabilities, raising security concerns

LiteLLM — Critical bytecode rewriting RCE (CVE-2026-40217)

LiteLLM versions through 2026-04-08 contain a critical remote code execution vulnerability via bytecode rewriting at /guardrails/test_custom_code endpoint.

lollms — Critical stored XSS in social features (CVE-2026-1115)

parisneo/lollms contains a critical stored XSS vulnerability (CVSS 9.6) allowing authenticated attackers to inject persistent malicious scripts through social posting features.

Anthropic — Mythos Preview autonomously discovers thousands of zero-day vulnerabilities across major systems

ModelContextProtocol — Java SDK DNS rebinding vulnerability allows MCP server takeover (CVE-2026-35568)

CVE-2026-35568 exposes DNS rebinding vulnerability in MCP Java SDK allowing attackers to bypass origin validation and execute arbitrary tool calls on local MCP servers with CVSS 7.6 severity.

Salt Security — 1H 2026 State of AI and API Security Report reveals agentic visibility crisis

Microsoft — CVE-2026-26113/26110 Preview Pane RCE

Microsoft Patch Tuesday addresses critical remote code execution vulnerabilities CVE-2026-26113 and CVE-2026-26110 affecting Office applications through Preview Pane exploitation.

KuCoin — AI trading agent vulnerabilities cause $45M crypto breaches

Protocol-level vulnerabilities in AI trading agents led to over $45M in crypto losses, exposing memory poisoning risks and weak authentication in autonomous trading systems.

Mercor — LiteLLM supply chain breach exposes 4TB of AI training data

AI data startup Mercor confirms 4TB breach from LiteLLM supply chain attack, exposing Meta, OpenAI, and Anthropic training methodologies and triggering class action lawsuit.

Microsoft — Agent Governance Toolkit addresses OWASP AI agent security risks

Microsoft releases open-source Agent Governance Toolkit addressing all 10 OWASP Agentic AI Top 10 risks with sub-millisecond policy enforcement for autonomous AI agents.

Anthropic — Claude discovers FreeBSD kernel RCE, writes full exploit

Nicholas Carlini (Anthropic) used Claude to discover CVE-2026-33017 in FreeBSD RPCsec GSS and autonomously develop a complete remote kernel exploit with root shell access.