Irregular: The first frontier security lab for AI

Kudos to Irregular (formerly Pattern Labs) for providing tools to help with the big problem of abuse of AI/LLM for cyberattacks. It may not help against nation-state attackers, who can roll out their own AI/LLMs, but it can definitely help against attackers abusing AI/LLM services.

Irregular (formerly Pattern Labs) is building the world’s first frontier security lab — with a singular mission: protecting the world in the time of increasingly capable and sophisticated AI systems. Today, we’re excited to be leading Irregular’s $80M Series A to accelerate that mission. The vision is led by Dan Lahav and Omer Nevo, two founders whose careers reflect a rare mix of deep technical rigor and applied impact. Together, they are assembling a world-class team to tackle one of the defining challenges of our time: how to secure, stress-test, and safeguard frontier AI systems before they’re deployed at scale. Irregular’s momentum is already evident — drawing top-tier technical talent and early recognition from industry and research partners. Trusted by Google, OpenAI, and Anthropic, the team is building practical, real-world security frameworks to safeguard frontier AI. Thrilled to be on this journey with Dan, Omer, and the team as they build the lab that will set the standard for AI safety and resilience. Shoutout to Sai Senthilkumar, Scott Raney and Jordan Segall for leading this investment.

Last night I was extremely fortunate to attend the Mindstone Toronto AI meetup .The Conference was a goldmine of critical insights, especially around AI security and MLOps.   Huge thanks to Himanshu J. for his incredibly timely and insightful session on the security aspects of AI at the Mindstone AI Conference yesterday. I especially valued his clear breakdown of how security challenges can be proactively dealt with—it's a critical discussion the entire industry needs to be having. His perspective on the vulnerabilities covering the AI space was truly eye-opening. Essential food for thought for anyone building or deploying AI.   I was equally impressed by the session of Brad Micklea. His presentation on KITOPS open-source packaging and versioning solution for AI/ML projects,  was incredibly eye-opening. The KITOPS solution—packaging complete ML projects into tamper-proof, cryptographically signed OCI-compliant artifacts—is exactly the secure gate we need between development and production.   . Great to see such focused, solution-oriented session ! #AICheck #MLOps #AISecurity #Jozu #MindstoneConference

Threat modeling is too often a high-effort ceremony instead of a low-friction, frequent habit. This post explores how leveraging AI and the command line can transform this critical security practice into a continuous and accessible process. I created a simple Gemini CLI extension to threat model nearly anything, from a simple concept to a complex system. The full blog post is here: https://lnkd.in/enygzFCG

Reality Defender's Ben Colman contributed to the World Economic Forum's new Responsible AI playbook, developed with input from dozens of global stakeholders. The findings are sobering: less than 1% of organizations have achieved full responsible AI implementation, despite widespread awareness of risks. This implementation gap becomes particularly acute when considering real-time threats like deepfake fraud and voice synthesis attacks that our platform addresses daily. The playbook provides nine actionable strategies across governance, development, and strategic dimensions, offering practical solutions to challenges including fragmented regulatory frameworks, unclear accountability structures, and legacy authentication systems inadequate for current AI threats. For enterprises seeking to maintain competitive velocity while implementing robust AI security measures, this framework represents a critical resource for transforming responsible AI from compliance overhead into operational advantage. https://lnkd.in/gx_crfpe

Great engineers are always thinking of securing their systems. But in an age where AI is everywhere, people don’t think twice about dumping huge contents of sensitive data into LLMs. At Scout, AI security isn’t just a checkbox, it’s woven into the fabric of the company and verified by our very own Allie Howe , vCISO. We spent the day at the AI Security Summit, learning from experts about the latest in what to be considering and always staying on top of this. #AI #AISecurity

Vulnerability management vendors promised prioritization with dashboards and reports. The reality? More noise, no context. Matt Alderman, CPO of CyberSaint, explains how we combine graph neural networks and LLMs to: 🔹 Connect threats → vulnerabilities → mitigating controls 🔹 Overlay real-time feeds for richer context 🔹 Run analytical models to quantify the impact of findings This isn’t “just another database or report.” It’s AI delivering meaningful, prioritized, and contextualized findings — turning raw data into purposeful value for decision-makers. Learn more about contextualized findings: https://lnkd.in/dAd-cT8X #CyberRiskManagement #AIinCyber #CyberGRC #RiskPrioritization

Everyone's talking about AI, but most enterprises aren't equipped for the new risks it creates. Data security is the top barrier to GenAI adoption. ePlus inc. and Palo Alto Networks outline 3 principles to protect your organization in the AI era. Start developing your foundation—dive into the guide: https://brnw.ch/21wWsqM

🚨🚨🚨 AI-Powered Attacks Are No Longer a Future Threat, They’re Here🚨🚨🚨 A recent Gartner study found that 29% of organizations have already faced attacks on their AI infrastructure — deepfakes, prompt manipulation, voice impersonation, and more. What this means: our defenses can’t stay static. Traditional tools alone are no longer enough. Here’s what security teams should focus on now: -Contextual anomaly detection — monitor behavior, not just signatures -Zero Trust + least privilege, even for AI/agent systems -Continuous validation of prompts and inputs to AI systems -Cross-domain collaboration — security, data science, devops must work tightly 🔍 If you’re in a security role: Which new attack vector involving AI worries you most today? Drop your thoughts below — let’s learn together. https://lnkd.in/eVwvfkrZ

Your new AI capabilities? They've just become your newest attack vector. Dr. Nimrod Partush, VP of AI and Innovation CYE, maps out the 4 most critical GenAI threats in every CISO's suite: https://lnkd.in/dkH3C5UA