Inference — Thursday, May 7, 2026

Thursday, May 7, 2026

New Releases

CrowdStrike Adds AI Model Scanning and Shadow AI Detection

globalsecuritymag.com · May 7

CrowdStrike shipped AI Model Scanning and Shadow AI detection inside Falcon, targeting cloud-hosted AI workloads and unsanctioned model use as new attack surfaces.

Bottom line

CrowdStrike is repositioning Falcon as the enforcement layer for AI supply chain security, not just endpoint and cloud workloads.

CrowdStrike Launches Falcon OverWatch for Microsoft Defender Customers

channelpostmea.com · May 7

CrowdStrike extended its managed threat hunting service to organizations running Microsoft Defender as their primary endpoint agent, without requiring a Falcon sensor swap.

Bottom line

CrowdStrike is monetizing Microsoft's endpoint market share rather than fighting it, turning Defender deployments into a top-of-funnel for managed services.

MongoDB Ships Unified AI Data Platform With Native Agent Memory

prnewswire.com · May 7

MongoDB announced at its London event today that persistent agent memory, vector search, automated Voyage AI embeddings, and a reranker model are now integrated into a single Atlas platform.

Bottom line

MongoDB is collapsing the four-to-six component RAG stack into one managed service, which directly challenges Pinecone, Weaviate, and standalone embedding providers.

Palo Alto PAN-OS Zero-Day Exploited Since April 9 by State Actors

bleepingcomputer.com · May 7

Palo Alto Networks disclosed CVE-2026-0300 today, a critical PAN-OS buffer overflow in the User-ID Authentication Portal that has been exploited in the wild since April 9 by suspected state-sponsored actors.

Bottom line

Nearly four weeks of undetected exploitation on perimeter firewalls underscores that state actors are increasingly targeting security infrastructure itself as the highest-value initial access vector.

PanDev Survey: Claude Code Saves 54 Minutes Daily vs. Copilot's 28

pandev-metrics.com · May 7

A 112-engineer, 14-team B2B survey measuring IDE heartbeat and session data in Q1 2026 found Claude Code users averaged 54 minutes of saved time per day versus 28 minutes for GitHub Copilot users.

Bottom line

The 26-minute daily gap compounds to roughly 90 hours per engineer per year, which at senior engineering fully-loaded costs makes the tool selection decision a budget-line item, not a preference.

Funding

Anthropic Secures SpaceX Colossus Compute Capacity in Infrastructure Deal May 7

Anthropic signed an agreement with SpaceX to access Colossus 1, the Memphis data center expected to exceed 220,000 Nvidia GPUs within a month, while simultaneously doubling Claude Code five-hour rate limits for Pro and Enterprise tiers — a move that signals compute supply is now Anthropic's primary constraint on agentic workload growth, not model capability.

CAISI Signs Pre-Deployment AI Vetting Deals With Microsoft, Google, xAI May 7

The US Department of Commerce's CAISI extended formal pre-deployment evaluation agreements to Microsoft, Google DeepMind, and xAI — joining Anthropic and OpenAI — establishing a five-lab government vetting pipeline that could become the de facto compliance gate for frontier model releases.