📰 Big News! Zscaler Acquires AI Security Pioneer @SplxAI! → bit.ly/4qDV4RV 💡 Why does this matter? AI innovations are reshaping industries, but securing the AI lifecycle—from development to deployment—is key to maximizing its impact. With @SplxAI joining forces with Zscaler, our Zero Trust Exchange now offers: ✅ Advanced AI Runtime Guardrails to protect sensitive data and block malicious prompts ✅ Proactive AI Asset Discovery to uncover risks in workflows, models, and deployments ✅ Automated Red Teaming to simulate attacks and fix vulnerabilities in real time ✅ Robust Governance & Compliance to secure AI investments at every stage 💡 Why should customers care? As AI drives adoption at breakneck speed, Zscaler’s newly combined security capabilities ensure organizations can innovate safely while mitigating risks. Together, we’re not just protecting AI—we’re empowering businesses to embrace its potential with trust, reliability, and unparalleled security. #AI #ZeroTrust #CyberSecurity #AILeadership

Join our founder and CTO Ante and Learn how to test & secure agentic AI systems in our hands-on workshop at @BSidesFRA: Challenges When Automating Agentic AI Red Teaming Sign up 👉 eventbrite.de/e/bsidesfrankf… See you there! #AIsecurity #RedTeaming #BSides

𝗕𝗹𝗮𝗰𝗸 𝗛𝗮𝘁 𝗨𝗦𝗔: 𝗜𝘁'𝘀 𝗮𝗹𝗺𝗼𝘀𝘁 𝘁𝗵𝗮𝘁 𝘁𝗶𝗺𝗲 𝗮𝗴𝗮𝗶𝗻... 🖤🧑‍💻 For the first time ever, @SplxAI is hitting the @BlackHatEvents show floor – and we’re bringing the most advanced 𝗣𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗳𝗼𝗿 𝗦𝗲𝗰𝘂𝗿𝗶𝗻𝗴 𝗔𝗜 with us. Our team has been shipping at light speed, and we can't wait to showcase our latest features that help secure AI apps and agents end-to-end – before deployment and during runtime. 📍 Find us at Booth #𝟲𝟰𝟮𝟯 in the Startup Zone 📅 Book a 𝟭:𝟭 𝘄𝗶𝘁𝗵 𝗼𝘂𝗿 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝘃𝗲 𝘁𝗲𝗮𝗺 or request an invite to our 𝗲𝘅𝗰𝗹𝘂𝘀𝗶𝘃𝗲 𝗖𝗜𝗦𝗢 𝗱𝗶𝗻𝗻𝗲𝗿: splx.ai/blackhat-2025 🚀 In town early? Come say hi at @BSidesLV! Let's continue making AI safe and trustworthy – together. 𝗦𝗲𝗲 𝘆𝗼𝘂 𝗶𝗻 𝗩𝗲𝗴𝗮𝘀. #BlackHat2025 #BSidesLV #SecurityforAI #AISecurity #LLMSecurity #SplxAI

🛠️ 𝗙𝗿𝗲𝗲 𝗧𝗼𝗼𝗹 𝗗𝗿𝗼𝗽: 𝗩𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗲 𝘆𝗼𝘂𝗿 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 𝗶𝗻 𝘀𝗲𝗰𝗼𝗻𝗱𝘀. As multi-agent systems become more complex – with MCP servers, A2A links, and toolchains across CrewAI, LangGraph, OpenAI Agents, AutoGen, or n8n frameworks – things start to become messy. This leads to security blind spots, logic errors, & hidden risks. Agentic Visualizer lets you upload your GitHub repo or source code ZIP-file and instantly see: → 𝗔𝗴𝗲𝗻𝘁𝘀, 𝘁𝗼𝗼𝗹𝘀, 𝗠𝗖𝗣 & 𝗔𝟮𝗔 𝗽𝗮𝘁𝗵𝘀 → 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻/𝗱𝗮𝘁𝗮 𝗳𝗹𝗼𝘄 𝗺𝗮𝗽𝘀 → 𝗣𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀 𝗹𝗼𝗴𝗶𝗰 𝘁𝗿𝗮𝗽𝘀 Try it now for free 👉 agentic-visualizer.splx.ai/

𝗚𝗣𝗧-𝟰𝗼 𝘃𝘀. 𝗼𝟯-𝗽𝗿𝗼: 𝗜𝘀 𝗺𝗼𝗿𝗲 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗮𝗹𝘄𝗮𝘆𝘀 𝗯𝗲𝘁𝘁𝗲𝗿? @OpenAI's recently released o3-pro model was designed to handle complex tasks with its advanced reasoning capabilities – but does that mean it's the best suitable model for every enterprise use case? We ran a full benchmark with the @SplxAI Platform, where we compared o3-pro to GPT-4o in a realistic scenario: A virtual AI assistant that helps users choose the right insurance policy. 📊 Here are the results: • o3-pro is 14x more expensive • Has 4x more failed test cases • Took 66 sec to respond vs. 1.5 sec for GPT-4o • Has much higher latency from excess reasoning o3-pro shines in advanced complexity, but GPT-4o wins on cost, speed & security in real-world AI applications. View the result metrics in the table below, and check out the full article for a detailed breakdown of the benchmark: splx.ai/blog/openai-o3-pro-v…

🔥 𝗡𝗲𝘄 𝗙𝗲𝗮𝘁𝘂𝗿𝗲: 𝗟𝗟𝗠 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀 📊 Choosing the wrong LLM can lead to leaks, exploits & reputational damage. ⚡ @SplxAI now lets you benchmark & compare top LLMs under real attack scenarios. ✅ Multi-prompt configs (no, basic, hardened) 🎯 Tested for security, trust, alignment 📊 Side-by-side comparisons 📥 Request any LLM benchmark 🔁 Continuously updated scores 🛡️ Know which LLMs hold up – before you deploy. 👉 𝗟𝗲𝗮𝗿𝗻 𝗺𝗼𝗿𝗲: splx.ai/platform/llm-benchma… 📰 𝗣𝗿𝗲𝘀𝘀 𝗿𝗲𝗹𝗲𝗮𝘀𝗲: splx.ai/blog/splxai-adds-llm… #AIsecurity #LLMBenchmarks #GenAI #AISafety #RedTeaming #SplxAI

✨ New Agentic Radar Integration: Microsoft @pyautogen is now supported! Agentic Radar, our open-source transparency scanner for agentic systems, now supports scanning AutoGen workflows – Microsoft’s powerful framework for building LLM-powered collaborative agents. Scan @pyautogen workflows to: 🔎 Visualize multi-agent interactions 🛠️ Track tool usage & dependencies ⚠️ Detect infinite loops & misconfigurations 🛡️ Flag OWASP GenAI-aligned risks 🧠 Auto-harden prompts Explore the example: splx.ai/blog/scanning-autoge… Get started: github.com/splx-ai/agentic-r…

We're thrilled to welcome our new Silver Sponsor @SplxAI ! They offer Continuous Security Testing for AI Assistants—detect and remediate AI vulnerabilities before they cause harm. splx.ai #BSidesFrankfurt #CyberSecurity #AI #InfoSec #Sponsorship #AgenticAI #AppSec

🤖 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 𝗮𝗿𝗲 𝗿𝗲𝗱𝗲𝗳𝗶𝗻𝗶𝗻𝗴 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 – but they also introduce new risks. Over 20% of time in white-collar jobs is spent on repetitive tasks. Agentic AI assistants are changing this by automating updates, generating docs, and filing JIRA tickets – all triggered through a company's communication tools like Slack. But greater autonomy means greater risk: 🔓 Authorization hijacking 🌀 Agentic hallucinations 📡 Multi-agent trust exploitation 📂 Tool-based data leakage These systems interact across tools and agents – often without human oversight. We break it all down (with examples) in our latest whitepaper: “𝙏𝙝𝙚 𝘾𝙪𝙧𝙧𝙚𝙣𝙩 𝙎𝙩𝙖𝙩𝙚 𝙤𝙛 𝘼𝙜𝙚𝙣𝙩𝙞𝙘 𝘼𝙄 𝙍𝙚𝙙 𝙏𝙚𝙖𝙢𝙞𝙣𝙜” 📥 Download it here → splx.ai/resources/the-curren… #AgenticAI #RedTeaming #GenAI #LLM #AIsecurity #MAS #SplxAI #AIsafety #LLMSecurity

⚠️ Multi-turn prompt injections are the next big threat to Agentic AI Workflows ⚠️ When targeting 𝗠𝘂𝗹𝘁𝗶-𝗔𝗴𝗲𝗻𝘁 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 (𝗠𝗔𝗦), prompt injection techniques become more complex – unfolding over multiple steps, priming agents across the conversation, bypassing layered guardrails, and triggering downstream tools (like email or document generation). In Agentic AI workflows, memory and context persist through multiple agents – making it easier for malicious payloads to slip through unnoticed and execute within the system. We break down how these attacks work in our latest whitepaper: “𝗧𝗵𝗲 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗦𝘁𝗮𝘁𝗲 𝗼𝗳 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 𝗥𝗲𝗱 𝗧𝗲𝗮𝗺𝗶𝗻𝗴” 📥 Download it here → splx.ai/resources/the-curren…

We’re beyond excited to welcome Joseph Thacker (@rez0__ ) – aka 𝙏𝙝𝙚 𝙋𝙧𝙤𝙢𝙥𝙩 𝙁𝙖𝙩𝙝𝙚𝙧 – as a strategic advisor to @SplxAI! A true pioneer of AI Red Teaming, Joseph has looked at 20 AI security tools and decided to join forces with the one he believes in most. His expertise will help us build the world’s most powerful AI Security Platform – and enable organizations globally to adopt secure AI at scale. Welcome on board, Joseph! 🚀

💥OUT NOW: The Current State of Agentic AI Red Teaming 💥 As LLM-powered apps evolve into autonomous, multi-agent systems, red teaming practices must also adapt to secure increasingly complex risks. Our new whitepaper breaks down: 📊 Insights from 100s of AI red team assessments ⚠️ Risks like RAG poisoning, tool misuse & data leaks 🧩 Model Context Protocol (MCP) vulnerabilities 🧠 Tools for threat modeling agentic systems 🎯 Real-world prompt injection examples 💼 Why continuous AI red teaming matters for business 📥 Get your copy now → splx.ai/resources/the-curren… Huge thanks to all writers and contributors! @Stanford @owasp @LLM_Top10 @FraunhoferSIT @cloudsa

🏛️ 𝗦𝗽𝗹𝘅𝗔𝗜 𝗮𝘁 𝘁𝗵𝗲 𝗡𝗲𝘄 𝗬𝗼𝗿𝗸 𝗦𝘁𝗼𝗰𝗸 𝗘𝘅𝗰𝗵𝗮𝗻𝗴𝗲 🏛️ A big thank you to @FintechTvGlobal and host @TheNadjaAtwal for hosting our CEO @KristianKamber at the @NYSE ! In this exclusive interview, Kristian reflects on the @SplxAI journey – coming out of stealth just over a year ago to now empowering 𝗙𝗼𝗿𝘁𝘂𝗻𝗲 𝟱𝟬𝟬 enterprises to adopt and deploy secure, compliant, and trustworthy AI apps and agents at scale. We’re proud to lead the charge of the 𝗦𝗮𝗳𝗲 𝗔𝗜 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 – helping industries like 𝗳𝗶𝗻𝗮𝗻𝗰𝗲, 𝗶𝗻𝘀𝘂𝗿𝗮𝗻𝗰𝗲, 𝗮𝗻𝗱 𝗯𝗲𝘆𝗼𝗻𝗱 leverage the power of AI with confidence. 📺 Watch the full interview here: fintech.tv/the-ai-security-r…

✅ 𝗦𝗰𝗮𝗻 & 𝗛𝗮𝗿𝗱𝗲𝗻 𝗦𝘆𝘀𝘁𝗲𝗺 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝗶𝗻 𝗔𝗜 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 𝘄𝗶𝘁𝗵 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗮𝗱𝗮𝗿 ✅ We’re excited to roll out the latest update to 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗮𝗱𝗮𝗿, our open-source security scanner for agentic workflows – now with automatic system prompt detection and hardening across all agents. System prompts define how an AI agent thinks and acts. When they’re vague or poorly defined, they invite jailbreaks, misbehavior, and data leaks. With this upgrade, your agents become instantly more: 🟩 𝗦𝗲𝗰𝘂𝗿𝗲 🟦 𝗥𝗲𝗹𝗶𝗮𝗯𝗹𝗲 🟪 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗮𝗯𝗹𝗲 What’s included: 🧭 Visibility into agents’ current instructions ⚠️ Detection of weak or risky system prompts 🛠️ Auto-suggested hardening fixes 📉 Lower risk of jailbreaks, leaks, and unwanted behavior 🔓 100% open-source and easy to integrate – making secure & aligned agentic workflows more accessible than ever. 🔗 Learn more: tinyurl.com/53svcdnv 🚀 Try it out: github.com/splx-ai/agentic-r…

Here is a 𝗠𝘂𝗹𝘁𝗶-𝗔𝗴𝗲𝗻𝘁 System Design for an 𝗔𝗜 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗥𝗲𝗱 𝗧𝗲𝗮𝗺𝗶𝗻𝗴 platform! Security of AI Systems remains one of the hard problems to be solved in the AI industry. The process of ensuring that your AI application can not be hacked is significantly different compared to regular software. I had a chance to sit down with experts in the field - splxAI - and chat about their platform implementation in detail. Check it out for free here: swrlai.com/4lTJnEa ✅ It is a Multi-Agent system worth studying for any 𝗔𝗜 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿. Here are the moving pieces: 𝟭. At the core we have the 𝗔𝘁𝘁𝗮𝗰𝗸 𝗔𝗴𝗲𝗻𝘁 that is supported and guided with context by other data sources and Agents. ℹ️ The resoning capabilities of this Agent are powered by a “Jailbroken LLM” which is capable of generating attacks without any ethical limitations. 𝟮. 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗮𝗱𝗮𝗿 is an agentic application that probes your target AI application code and discovers details to help guide the Attack Agent in the right direction. 𝟯. Additional data provided by the user that is testing their application - Domain Details, RAG Examples etc. 𝟰. 𝗥𝗲𝗱 𝗧𝗲𝗮𝗺 that curates the database of 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝘁𝘁𝗮𝗰𝗸𝘀 and keeps it up to date with the latest attack patterns that have proven successful for different types of Agentic Applications. 𝟱. 𝗖𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝗰𝗲 𝗔𝗴𝗲𝗻𝘁 that continuously checks that the latest compliance standards are properly mapped to foundational attacks workflows. 𝟲. Once attack attempts are completed, the Attack Agent stores the conversation logs for further analysis. 𝟳. 𝗗𝗲𝘁𝗲𝗰𝘁𝗼𝗿 𝗔𝗴𝗲𝗻𝘁 works in conjunction with Attack Agent to understand if target of attack is achieved shapes feedback on attack results as a readable information for red teamers 𝟴. A 𝗥𝗲𝗺𝗲𝗱𝗶𝗮𝘁𝗶𝗼𝗻 𝗔𝗴𝗲𝗻𝘁 that analyses attack results, groups them and simplifies remediation steps that are then suggested to the user. Thank you @SplxAI for collaborating on this post and helping to educate the community! What is your experience with Red Teaming your AI Systems? Let me know in the comments 👇 #LLM #AI #MachineLearning

Red teaming your Gen AI applications with @SplxAI - field report from @OpenAI hackathon in Warsaw @AITinkerers

🧠 𝗚𝗣𝗧-𝟰.𝟭 𝗶𝘀 𝗵𝗲𝗿𝗲 – 𝗕𝘂𝘁 𝗵𝗼𝘄 𝗦𝗮𝗳𝗲 𝗶𝘀 𝗶𝘁 𝗳𝗼𝗿 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝗔𝗽𝗽𝘀? 💥 @OpenAI released the new GPT-4.1 model family without a safety report. Most organizations will likely switch from GPT-4o to the new model without proper due diligence and testing. To evaluate security and safety implications of GPT-4.1, we ran a red teaming assessment with 1,000 simulated interactions. The results were concerning: ❗ 𝗚𝗣𝗧-𝟰.𝟭 𝗮𝗹𝗹𝗼𝘄𝗲𝗱 𝟯𝘅 𝗺𝗼𝗿𝗲 𝗺𝗶𝘀𝘂𝘀𝗲 𝘁𝗵𝗮𝗻 𝗚𝗣𝗧-𝟰𝗼 ❗ 𝗦𝘄𝗶𝘁𝗰𝗵𝗶𝗻𝗴 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘂𝗽𝗱𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 𝗽𝗿𝗼𝗺𝗽𝘁 𝗱𝗼𝘂𝗯𝗹𝗲𝘀 𝘁𝗵𝗲 𝗿𝗶𝘀𝗸 ❗ 𝗘𝘃𝗲𝗻 𝘂𝘀𝗶𝗻𝗴 𝗢𝗽𝗲𝗻𝗔𝗜’𝘀 𝗽𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 𝗴𝘂𝗶𝗱𝗲𝗹𝗶𝗻𝗲𝘀 𝗶𝗻𝗰𝗿𝗲𝗮𝘀𝗲𝗱 𝘃𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 💡 If you're upgrading your AI apps to GPT-4.1 – don’t do it blindly. Every model change requires a re-evaluation of security, safety, and system prompts. 📄 Read the full research report: splx.ai/blog/the-missing-gpt…

🔎 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗮𝗱𝗮𝗿 𝗻𝗼𝘄 𝗱𝗲𝘁𝗲𝗰𝘁𝘀 𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹 (@MCP_Community) 𝘀𝗲𝗿𝘃𝗲𝗿 𝗰𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻𝘀 𝗶𝗻 𝗢𝗽𝗲𝗻𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 – giving you visibility into how agents access external systems and helping you secure data flows. Why does this matter? 🔍 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆 – Know which external systems agents access 🛡️ 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 – Spot risky connections & enforce least privilege 📊 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 – Track how sensitive data flows 📖 Read Our Example: splx.ai/blog/agentic-radar-n… 🔗 Get Started on GitHub: github.com/splx-ai/agentic-r… #AgenticAI #AIsecurity #MCP #OpenAI #GenAI #AgenticRadar #AItransparency #RedTeaming