TCG

RSAC is hosting my talk "Securing Federal LLM Systems: Multi-Agent Workflows and Data Safety" on the "OneRSAC Portal" at https://lnkd.in/gbihM3pK . (They are currently offering access to the full Portal for free, but will eventually limit library access to "Premium" members.) Out of curiosity, I pressed their "Summarize with AI" button. Here's their AI-generated summary: SUMMARY Robert Buccigrossi discusses securing federal large language model systems, focusing on compliance, infrastructure, and the risks associated with agentic AI. He highlights the importance of using role-based retrieval augmented generation to protect sensitive information and outlines tools and methods for implementing safety measures in federal AI environments. IDEAS: Infrastructure and Compliance - Use FedRAMP-compliant services like Azure Government, AWS GovCloud, and Google Gemini for federal large language models. - The choice between cloud and on-premises systems can significantly impact cost and efficiency. - FedRAMP High systems are crucial for meeting federal security standards. Risks of Data Handling - Fine-tuning large language models with sensitive data carries a risk of data memorization, which can be exploited via prompt injection attacks. - Multi-agent systems can help, but they are not fully reliable against vulnerabilities like jailbreaking. - Importance of not exposing LLMs to sensitive data to mitigate risks of insider and outsider threats. Methods of Improvement - Role-based retrieval augmented generation (RAG) can limit user access to sensitive information based on their roles. - Parallel processing in multi-agent systems can optimize interactions and enhance security protocols. - Implementation of a robust framework and auditing system is essential to keep track of interactions and secure sensitive information. Recommendations for Practice - Agencies should evaluate their current systems to incorporate secure measures such as role-based RAG. - Continuous monitoring and logging of interactions with LLMs are crucial for identifying potential data leaks. - Engage in collaborative efforts to share knowledge about AI safety standards across agencies. QUOTES: "Never give a large language model data that you don't want the user to access." "It's a cat and mouse game where there's prompt engineering, there's jailbreaking." "The economy of the cloud versus on-prem is kind of upside down with large language models." "Data memorization...is a risk in that it can be extracted with sufficiently savvy attacks." "Role-based retrieval augmented generation...avoids those potential data leaks."

Is AI a cheat code, or a necessary skill? I recently explored this question in a blog post, and what I found might surprise you. AI is rapidly transforming the developer landscape. I dive into why simply using AI isn't enough – it's about mastering best practices to truly leverage its power. It's not just about speed; it's about integration, compliance, and delivering real value. What are your thoughts on AI as a core skill for professionals today? Share in the comments! Read the full article in the first comment below! 👇 #AI #ArtificialIntelligence #TechSkills #SoftwareDevelopment #Innovation #TCG

  We're proud to announce that TCG has, once again, certified CMMI Level 3 for both service and development! This milestone underscores our commitment to high-quality solutions and process improvement. A huge thank you to our dedicated team for making this possible. Together, we look forward to continuing our journey of excellence and innovation! #CMMI #TeamSuccess #Innovation #TCG

📼 The Geek Within REWIND Mondays! Year 2020 In episode #8 of The Geek Within, my guest was Michael Drescher, who was a Business Development Executive at TCG and is now a Federal Growth Executive at TCG. We hear how he got into IT and his thoughts on the future. 🔊 Listen to episode 8: Spotify - https://lnkd.in/eVzH2RVc YouTube - https://lnkd.in/eS9y5-Qs 🔗 You can engage Michael here: LinkedIn - https://lnkd.in/e95pnx9Z 🎵 Other audio platforms: Apple Podcasts - https://lnkd.in/e78b2F4J Amazon Music - https://lnkd.in/edy28nFG iHeartRadio - https://lnkd.in/eP5n-8dr Consider dropping a review on the podcast platform of your choice!

Really appreciated the conversations at the Engage FedGov: Advancing Federal Innovation – AI, Acquisition, and Strategy event at the Samuel Riggs IV Alumni Center on friday. Lots of thoughtful dialogue and plenty to reflect on. I had a great time learning more with Tige Nishimoto, PMP, PMI-ACP, CSM, SPC and I am excited to bring back what I learned to TCG! #EngageFedGov #Innovation #PublicSector #Strategy #EmergingTech