TLDR IT 2026-06-08

A critical PAN-OS vulnerability affecting GlobalProtect portals and gateways is being used to bypass authentication and establish unauthorized VPN access. CISA added CVE-2026-0257 to its Known Exploited Vulnerabilities catalog, and Palo Alto has issued patches across affected PAN-OS versions.

Enterprises are deploying agents faster than they are adding enforceable governance, monitoring, and identity controls. Deloitte found that only 21% of organizations have mature governance for autonomous AI agents, while 73% are concerned about AI security and privacy risks.

Enterprises are not abandoning AI, but they are starting to cut pilots that lack clear ROI, sustainable costs, or manageable risk. Workflow-specific AI tools and narrowly scoped agents are more likely to survive than broad per-seat copilots with vague productivity claims.

Many channel partners are not ready to support enterprise AI moving from pilots to production. The bottlenecks are increasingly around storage, data architecture, cloud costs, power, cooling, and hybrid infrastructure rather than the AI models themselves.

Stop burning tokens trying to teach AI your business. Instead, give your agents the context they need with Celonis. The Celonis Context Model spins up a digital twin of your organization that agents draw on to reason through business challenges. That's the clarity you're looking for. Learn how Celonis works

Google introduces an agentic RAG framework for dependable enterprise question answering across multiple data sources. Integrated into the Gemini Enterprise Agent Platform, it uses specialized agents to iteratively retrieve until the needed information is present before generating a grounded response. In evaluations on FramesQA and internal datasets, it improves factuality accuracy by up to 34% and achieves high correctness in cross-corpus retrieval while keeping latency similar.

Snowflake is expanding Claude access across Cortex AI, Cortex Code, and Snowflake Intelligence so enterprises can build agents and apps on governed internal data. The partnership targets “pilot purgatory,” where AI projects stall because of integration, security, privacy, and compliance barriers.

GitHub Copilot plans now use GitHub AI Credits instead of premium request counts. Usage is based on token consumption, including input, output, and cached tokens, giving admins more direct visibility into AI coding costs.

AI-built apps differ from traditionally engineered software because generated code is untrusted by default and introduces new security and infrastructure risks in production. Safe deployment requires microVM sandbox isolation, secrets management, RBAC, CI/CD with per-PR preview environments, and support for BYOC and GPU workloads. Many fast deployment options fall short on isolation and production-grade infrastructure needed to run AI-generated apps securely.

GitHub Copilot now bills all plans based on GitHub AI Credits consumed, giving IT and engineering admins a more direct way to track AI usage costs.

Mandiant says the group is using social engineering to access document management and cloud storage systems before rapidly exfiltrating sensitive legal and financial files.