Research, red-teaming insights, and AI security news from Adversa AI — covering agentic AI, GenAI, LLM, and MCP security.https://adversa.ai/https://adversa.ai/blog/top-mcp-security-resources-august-2026/https://adversa.ai/blog/top-mcp-security-resources-august-2026/August 2026 MCP security roundup: the July 28 spec revision hardens authorization, a confused deputy hijacks Azure DevOps review agents, and scanners arrive.Thu, 06 Aug 2026 12:50:00 GMThttps://adversa.ai/blog/ai-coding-agent-incidents/https://adversa.ai/blog/ai-coding-agent-incidents/Nine documented AI coding agent incidents, from Cursor and Gemini CLI to Replit, Kiro and Claude Opus 5. What each agent did and why the guardrails failed.Tue, 04 Aug 2026 10:57:00 GMThttps://adversa.ai/blog/top-ai-coding-agent-security-resources-august-2026/https://adversa.ai/blog/top-ai-coding-agent-security-resources-august-2026/August 2026's AI coding agent security roundup: a symlink flaw across six assistants, Cursor's CVSS 9.8 pair, a private repo leak, and a wiped database.Mon, 03 Aug 2026 13:24:00 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-august-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-august-2026/Four teams broke production AI agents in ten days. Twenty resources on the Kiro RCE, Cursor's CVSS 9.8 pair, memory poisoning, and sandbox escapes.Fri, 31 Jul 2026 14:47:00 GMThttps://adversa.ai/blog/agent-skill-scanners-bypass-eight-tested/https://adversa.ai/blog/agent-skill-scanners-bypass-eight-tested/We ran eight open source AI skill scanners against real attacks. A malicious skill got past all eight, including the current OASB leaderboard leader.Thu, 30 Jul 2026 18:45:00 GMThttps://adversa.ai/blog/openai-ai-agent-sandbox-escape-hugging-face-breach/https://adversa.ai/blog/openai-ai-agent-sandbox-escape-hugging-face-breach/An OpenAI benchmark agent escaped its sandbox and breached Hugging Face, logging 17,000+ actions. What the evidence shows, what's unproven, and what to fix.Mon, 27 Jul 2026 00:00:00 GMThttps://adversa.ai/blog/github-prodbot-ctf-ai-red-teaming-agent/https://adversa.ai/blog/github-prodbot-ctf-ai-red-teaming-agent/How our AI red teaming agent solved all five levels of GitHub's ProdBot challenge, one distinct exploit per level, and what defenders can take from it.Fri, 17 Jul 2026 06:40:38 GMThttps://adversa.ai/blog/cursor-security-deepjack-deeplink-vulnerability-mcp-rce/https://adversa.ai/blog/cursor-security-deepjack-deeplink-vulnerability-mcp-rce/A crafted cursor:// link installs an attacker-controlled MCP server that runs unsandboxed commands under your account, and the install dialog hides the payload.Wed, 15 Jul 2026 12:59:07 GMThttps://adversa.ai/blog/top-ai-coding-agent-security-resources-july-2026/https://adversa.ai/blog/top-ai-coding-agent-security-resources-july-2026/This month exposed a harsh reality: autonomous coding agents are expanding the enterprise attack surface faster than we can patch them.Thu, 09 Jul 2026 11:33:53 GMThttps://adversa.ai/blog/top-mcp-security-resources-july-2026/https://adversa.ai/blog/top-mcp-security-resources-july-2026/July's security digest covers the critical MCP vulnerabilities, real-world MCP exploitation, NSA's official MCP hardening guidelines.Mon, 06 Jul 2026 14:44:43 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-july-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-july-2026/July 2026's agentic AI security roundup: agentic zero trust whitepapers, AutoJack & other new exploits, and the newest agent defenses.Thu, 02 Jul 2026 13:46:26 GMThttps://adversa.ai/blog/opensource-ai-coding-agents-shell-injection-vulnerability/https://adversa.ai/blog/opensource-ai-coding-agents-shell-injection-vulnerability/AI coding agents run shell commands with your full account authority. GuardFall is a shell injection class that walks straight past their command filters.Tue, 30 Jun 2026 13:00:59 GMThttps://adversa.ai/blog/owasp-asi03-identity-privilege-abuse-in-ai-agents/https://adversa.ai/blog/owasp-asi03-identity-privilege-abuse-in-ai-agents/OWASP ranks Identity & Privilege Abuse #3 because it sets the blast radius for every other AI agent risk.Thu, 25 Jun 2026 09:43:35 GMThttps://adversa.ai/blog/solving-defcon-ai-ctf-with-ai-red-teaming-agent/https://adversa.ai/blog/solving-defcon-ai-ctf-with-ai-red-teaming-agent/Every DEFCON Singapore AI CTF task dissected. Learn how our AI red teaming agent used transform-based LLM jailbreak techniques to bypass guardrails.Tue, 23 Jun 2026 08:45:00 GMThttps://adversa.ai/blog/top-genai-security-resources-june-2026/https://adversa.ai/blog/top-genai-security-resources-june-2026/June 2026 in GenAI security: jailbreaks that will not die, ChatGPhish turning page summaries into phishing, and 1,652 wide-open Ollama APIs.Mon, 08 Jun 2026 05:01:40 GMThttps://adversa.ai/blog/adversa-ai-launches-airq-framework-report/https://adversa.ai/blog/adversa-ai-launches-airq-framework-report/Open-source methodology by OWASP, CoSAI, CSA, and NIST contributors enables enterprises to compare AI agents by security risk for the first timeThu, 04 Jun 2026 13:00:23 GMThttps://adversa.ai/blog/top-mcp-security-resources-june-2026/https://adversa.ai/blog/top-mcp-security-resources-june-2026/June 2026 in MCP security: Censys counted 12,520 Internet-accessible MCP services, most of them unauthenticated, and the measurements got a lot more concrete.Thu, 04 Jun 2026 04:59:57 GMThttps://adversa.ai/blog/airq-ai-risk-quadrant-for-agents-top-100-agents-scored-for-attack-defense-blast-radius/https://adversa.ai/blog/airq-ai-risk-quadrant-for-agents-top-100-agents-scored-for-attack-defense-blast-radius/The AIRQ report scores 100 AI agents on attack surface, blast radius, and defenses. The AIRQ framework lets you assess your own stack.Wed, 03 Jun 2026 13:00:54 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-june-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-june-2026/June 2026's agentic AI security roundup: coding agent RCEs, Microsoft Semantic Kernel flaws, a Copilot backdoor, and the newest agent defenses.Mon, 01 Jun 2026 06:00:02 GMThttps://adversa.ai/blog/the-approval-prompt-is-lying-to-you-symlink-rce-in-five-ai-coding-agents-claude-code-cursor-antigravity-copilot-grok-build/https://adversa.ai/blog/the-approval-prompt-is-lying-to-you-symlink-rce-in-five-ai-coding-agents-claude-code-cursor-antigravity-copilot-grok-build/A SymJack attack tricks AI coding assistants into RCE through a symlink-disguised file copy. We tested six major tools and every one of them was vulnerable.Tue, 26 May 2026 07:25:58 GMThttps://adversa.ai/blog/owasp-asi02-tool-misuse-and-exploitation-the-definitive-security-guide/https://adversa.ai/blog/owasp-asi02-tool-misuse-and-exploitation-the-definitive-security-guide/OWASP ASI02 - full technical guide. How AI agents misuse legitimate tools to wipe drives and delete databases. Plus mitigations and controls that stop it.Mon, 18 May 2026 06:02:12 GMThttps://adversa.ai/blog/ai-risk-management-insurance-what-the-new-exclusions-mean/https://adversa.ai/blog/ai-risk-management-insurance-what-the-new-exclusions-mean/Major insurers are adding AI-related exclusions to their policies.Wed, 13 May 2026 11:48:14 GMThttps://adversa.ai/blog/top-genai-security-resources-may-2026/https://adversa.ai/blog/top-genai-security-resources-may-2026/May 2026 in GenAI security: Anthropic's Mythos model running a 32-step network attack, new research bypassing GPT-5.4 safety, and LLM-specific CIS guidance.Mon, 11 May 2026 06:00:39 GMThttps://adversa.ai/blog/trustfall-coding-agent-security-flaw-rce-claude-cursor-gemini-cli-copilot/https://adversa.ai/blog/trustfall-coding-agent-security-flaw-rce-claude-cursor-gemini-cli-copilot/A Claude Code trust dialog regression and a settings scope inconsistency let a cloned repo run unsandboxed code with one keypress, and none on CI runners.Thu, 07 May 2026 13:00:08 GMThttps://adversa.ai/blog/top-mcp-security-resources-may-2026/https://adversa.ai/blog/top-mcp-security-resources-may-2026/May 2026 in MCP security: a CVSS 9.8 flaw in NGINX integrations, an STDIO transport vulnerability exposing hundreds of thousands of servers, and the response.Wed, 06 May 2026 08:55:09 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-may-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-may-2026/May 2026 in agentic AI security: 29 resources on in-the-wild exploits, the Claude Code source leak, and vendors quietly patching their autonomous agents.Mon, 04 May 2026 20:07:49 GMThttps://adversa.ai/blog/ai-driven-exploitation-mythos-what-comes-next/https://adversa.ai/blog/ai-driven-exploitation-mythos-what-comes-next/Anthropic's Mythos completed a 32-step network attack autonomously in hours.Thu, 30 Apr 2026 07:01:58 GMThttps://adversa.ai/blog/iicl-attack-gpt-5-4-safety-bypass-in-context-learning/https://adversa.ai/blog/iicl-attack-gpt-5-4-safety-bypass-in-context-learning/IICL is a structural jailbreak bypassing GPT-5.4 safety with just two words.Thu, 23 Apr 2026 11:58:40 GMThttps://adversa.ai/blog/manual-in-house-continuous-red-teaming-agentic-ai-coverage-cost/https://adversa.ai/blog/manual-in-house-continuous-red-teaming-agentic-ai-coverage-cost/A practical framework for comparing manual, in-house, and continuous red teaming of AI agents across coverage, cost, staffing, and compliance needs.Thu, 16 Apr 2026 10:19:13 GMThttps://adversa.ai/blog/adversa-ai-wins-artificial-intelligence-excellence-award-safety-alignment-category/https://adversa.ai/blog/adversa-ai-wins-artificial-intelligence-excellence-award-safety-alignment-category/Adversa AI won in the Safety and Alignment category, recognized for advancing real-world AI safety through continuous adversarial testing of AI systems.Fri, 10 Apr 2026 04:55:47 GMThttps://adversa.ai/blog/asi01-agent-goal-hijack-a-practical-security-guide/https://adversa.ai/blog/asi01-agent-goal-hijack-a-practical-security-guide/Full technical guide to Agent Goal Hijack, the #1 risk in the OWASP Agentic Top 10.Thu, 09 Apr 2026 05:02:34 GMThttps://adversa.ai/blog/top-genai-security-resources-april-2026/https://adversa.ai/blog/top-genai-security-resources-april-2026/April 2026 in GenAI security: 19 resources on the LiteLLM supply chain compromise, 128K context window poisoning, and compound RAG database exploits.Wed, 08 Apr 2026 05:22:36 GMThttps://adversa.ai/blog/top-mcp-security-resources-april-2026/https://adversa.ai/blog/top-mcp-security-resources-april-2026/Our April 2026 MCP resources digest highlights the latest vulnerability research and practical defenses.Mon, 06 Apr 2026 05:15:18 GMThttps://adversa.ai/blog/critical-claude-code-vulnerability-deny-rules-silently-bypassed-because-security-checks-cost-too-many-tokens/https://adversa.ai/blog/critical-claude-code-vulnerability-deny-rules-silently-bypassed-because-security-checks-cost-too-many-tokens/Adversa AI red team found Claude Code's deny rules silently stop working after 50 subcommands. The fix exists in Anthropic's codebase. They never shipped itThu, 02 Apr 2026 10:54:01 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-april-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-april-2026/April 2026 in agentic AI security: privilege escalation in OpenClaw, the hijacking of Chrome's Gemini Live assistant, and the RSA conversation around both.Wed, 01 Apr 2026 10:18:01 GMThttps://adversa.ai/blog/adversa-ai-wins-most-innovative-agentic-ai-security-at-global-infosec-awards-during-rsa-conference-2026/https://adversa.ai/blog/adversa-ai-wins-most-innovative-agentic-ai-security-at-global-infosec-awards-during-rsa-conference-2026/Recognized Among Hundreds of Vendors for Advancing Continuous AI Red Teaming and Agentic AI Security.Tue, 31 Mar 2026 06:50:16 GMThttps://adversa.ai/blog/agentic-ai-red-teaming-p3/https://adversa.ai/blog/agentic-ai-red-teaming-p3/The six threat actors your red team should be simulating, the five expertise domains needed to find them, and the uncomfortable math between the two.Mon, 30 Mar 2026 05:09:05 GMThttps://adversa.ai/blog/adversa-ai-agent-breaks-defenses-gandalf-agent-ctf-red-teaming/https://adversa.ai/blog/adversa-ai-agent-breaks-defenses-gandalf-agent-ctf-red-teaming/Our agent made it to the top 3 in Gandalf CTF for agents. It predicts vulnerabilities before sending a single attack.Tue, 24 Mar 2026 11:27:47 GMThttps://adversa.ai/blog/openclaw-enterprise-security-strategy/https://adversa.ai/blog/openclaw-enterprise-security-strategy/OpenClaw proved high-agency AI works, but banning it won't stop shadow AI or close the competitive gap.Fri, 20 Mar 2026 11:45:49 GMThttps://adversa.ai/blog/ai-guardrails-vs-ai-red-teaming/https://adversa.ai/blog/ai-guardrails-vs-ai-red-teaming/AI guardrails block known threats — but four attack patterns consistently bypass them.Thu, 19 Mar 2026 14:11:12 GMThttps://adversa.ai/blog/agentic-ai-red-teaming-p2/https://adversa.ai/blog/agentic-ai-red-teaming-p2/Most AI security assessments focus solely on prompt injection, leaving up to 90% of your agentic AI attack surface exposed.Mon, 16 Mar 2026 05:02:43 GMThttps://adversa.ai/blog/top-genai-security-resources-march-2026/https://adversa.ai/blog/top-genai-security-resources-march-2026/March’s GenAI security digest highlights a shift toward industrial-scale AI exploitation and systemic manipulation.Mon, 09 Mar 2026 07:32:18 GMThttps://adversa.ai/blog/top-mcp-security-resources-march-2026/https://adversa.ai/blog/top-mcp-security-resources-march-2026/March 2026 in MCP security: overthinking loops that drain API budgets, high-privilege RCE in Claude desktop extensions, and a widening MCP attack surface.Thu, 05 Mar 2026 09:47:18 GMThttps://adversa.ai/blog/top-agentic-ai-security-resources-march-2026/https://adversa.ai/blog/top-agentic-ai-security-resources-march-2026/March 2026 in agentic AI security: 27 resources on the OpenClaw rush, identity files like SOUL.md, memory poisoning, and GitHub Copilot vulnerabilities.Tue, 03 Mar 2026 12:40:25 GMThttps://adversa.ai/blog/openclaw-attacks-real-scenarios-owasp-mitre-csa-defense-guide/https://adversa.ai/blog/openclaw-attacks-real-scenarios-owasp-mitre-csa-defense-guide/In the past 30 days, MITRE, cybersecurity vendors, and independent researchers documented seven distinct attack paths against OpenClaw AI agents.Wed, 25 Feb 2026 06:00:08 GMThttps://adversa.ai/blog/openclaw-ai-agent-security-threats-mapped-owasp-mitre/https://adversa.ai/blog/openclaw-ai-agent-security-threats-mapped-owasp-mitre/OpenClaw AI agents sit on top of your files, credentials, and inbox.Thu, 19 Feb 2026 10:47:48 GMThttps://adversa.ai/blog/business-risks-of-agentic-ai-csuite-guide-to-owasp-agentic-top-10/https://adversa.ai/blog/business-risks-of-agentic-ai-csuite-guide-to-owasp-agentic-top-10/AI agents don’t just suggest transfers — they execute them.Wed, 18 Feb 2026 15:27:26 GMThttps://adversa.ai/blog/secureclaw-open-source-ai-agent-security-for-openclaw-aligned-with-owasp-mitre-frameworks/https://adversa.ai/blog/secureclaw-open-source-ai-agent-security-for-openclaw-aligned-with-owasp-mitre-frameworks/The OpenClaw story proves that agentic AI security will be more challenging than traditional application security.Tue, 17 Feb 2026 13:05:12 GMThttps://adversa.ai/blog/adversa-ai-launches-secureclaw-open-source-security-solution-for-openclaw-agents/https://adversa.ai/blog/adversa-ai-launches-secureclaw-open-source-security-solution-for-openclaw-agents/SecureClaw is an open source security suite for OpenClaw agents: 50 audit checks, automated hardening modules, and a defense layer for agentic deployments.Sun, 15 Feb 2026 11:11:39 GMThttps://adversa.ai/blog/unexpected-code-execution-in-agentic-ai-definitive-guide-to-asi05/https://adversa.ai/blog/unexpected-code-execution-in-agentic-ai-definitive-guide-to-asi05/A practical guide to unintended code execution in agentic AI: risk profiles, attack vectors, and concrete mitigations for agents that generate and run code.Thu, 12 Feb 2026 06:19:33 GMT