- Cloud Security Discussion Session
Quick recap. The Cloud Security Office Hours meeting featured discussions on recent AI model hacking incidents, the importance of query languages in security, and strategies for establishing security credibility. Shawn shared new how-to guides on the website for learning various query languages. The group debated the risks of AI models providing false information and the challenges of detecting such behavior. Neil raised the topic of how companies and individuals can build security credibility, with input from Matt and Jay on the importance of practitioner experience and tailoring the security message to different audiences. The conversation also touched on the impact of security incidents on vendor trust and the role of cyber insurance in evaluating tech stacks. Personal updates included Chris Richardson's return to Texas and Jay's recent vacation in Europe.
Cloud Security Discussion Session
Shawn welcomed Chris Richardson back to Texas and discussed his new role and travel plans. Jay returned from vacation and shared details about his trip to France and Italy. The meeting began with introductions and casual conversation, setting the stage for an open discussion session on cloud security topics. Jay mentioned recent news about model providers being hacked, specifically noting concerns about foreign models like GLM being potentially dangerous.
Query Languages Learning Resource
Shawn announced new how-to guides on query languages added to the website, designed as a cross-platform learning resource using CloudTrail exercises that work across nine different query languages including Datadog and Elastic. Neil confirmed the value of query language skills for working with log data, sharing his experience with Microsoft SQL Server. The discussion also touched on concerns about AI agents hacking online retailers and legal implications of such activities, though no specific decisions or action items were established regarding these topics.
Big Data Query Languages Discussion
The group discussed various query languages and tools for working with big data, including Kusto, JSON processing with JQ, and Python. Brian raised concerns about using AI tools like ChatGPT for learning these languages, questioning how to protect against potential misinformation or dangerous commands. The discussion also touched on minimizing container environments to prevent unauthorized tool usage and the challenges of working with restricted agents in professional settings.
AI Trust and Security Credibility
The group discussed challenges with AI model trust and security credibility. Brian raised concerns about detecting when AI models lie or hide their actions, while Matt and Neil shared experiences about evaluating AI responses requiring subject matter expertise. The conversation then shifted to how companies establish security credibility, with Jay explaining that credibility depends on the audience and noting that many CISOs struggle to communicate with executive leadership. The discussion concluded with reflections on vendor reliability and cyber insurance, including whether poor security performance by vendors might impact insurance premiums.