← All digests

🔎 Research Digest — 2026-08-14

Executive signal:

  • SharePoint CVE-2026-55040 has moved from patch guidance to active exploitation after a public PoC; any on-prem exposure is urgent.
  • Microsoft and OpenAI are both reframing agent adoption around economics: routing, memory, tool design, budgets, and latency now matter more than model hype.
  • Google is compressing agent economics further with Gemini 3.7 Flash, reinforcing that practical coding-agent costs are falling fast across vendors.
  • Gaming operators are pivoting from sweepstakes toward prediction-market rails; Fliff's filing is a concrete sign of where platform pressure is heading.

🎯 Today's Priority

💹 Markets & Macro

  • Fact: SPY closed at USD 777.88 (+0.70%), MSFT at USD 496.88 (+0.90%), and NVDA at USD 225.30 (+0.54%).
  • Fact: MGM closed at USD 44.07 (+0.20%) and CZR at USD 29.66 (+0.10%), so the gaming watchlist was basically flat to slightly firmer.
  • Fact: BTC was at USD 63,361.17 (-0.07%) and ETH at USD 1,883.21 (+0.27%) on the 2026-08-14 UTC capture.

🤖 AI & Agents

  • Fact: OpenAI says GPT-5.6 Luna reached 84.04% on BrowseComp for USD 1.33 versus GPT-5.5 at 84.36% for USD 33.27, and says retained reasoning plus compaction lifted ARC-AGI-3 from 13.3% to 38.3% while using roughly 6x fewer output tokens. Interpretation: agent-system design is now a first-order cost lever.
  • Fact: OpenAI says GPT-5.6 Sol on Ultrafast can run up to 14x faster than Standard and deliver up to 750 output tokens per second. Interpretation: incident response, live research, and support workflows are becoming more realistic frontier-model use cases.
  • Fact: Google says Gemini 3.7 Flash improved FrontierCode 1.1 Main to 43.6% from 34.4% for 3.6 Flash and is priced at USD 0.75 per 1M input tokens and USD 3.75 per 1M output tokens. Interpretation: the workhorse-agent price war is accelerating.

☁️ Cloud & 🛠️ DevOps

  • Fact: Microsoft says more than 100,000 organizations are building on Foundry and cites an IDC survey where 71% of business leaders plan to increase AI budgets. Interpretation: Azure's AI story is shifting from pilots to governed production spending.
  • Fact: Microsoft frames routing, caching, tool scoping, memory, and budget controls as the operating levers for agent cost. Interpretation: for Casper, the real AZ-104 lesson is that AI now sits inside normal cloud governance.

🔐 Cybersecurity

  • Fact: The Hacker News says exploitation of SharePoint CVE-2026-55040 picked up after Rapid7 released a PoC; Rapid7 says the bug lets an unauthenticated attacker forge a valid JWT and impersonate a SharePoint user or administrator. Interpretation: if any SharePoint farm still exists in Casper's vendor chain, July patch validation should move to the front of the queue.

Saved Knowledge / LLM Wiki Candidates

  • Created raw/articles/azure-agent-optimization-economics-2026-08-12.md
  • Created raw/articles/openai-gpt-5-6-builder-guide-2026-08-13.md
  • Created raw/articles/openai-gpt-5-6-ultrafast-2026-08-13.md
  • Created raw/articles/google-gemini-3-7-flash-2026-08-13.md
  • Created raw/articles/thn-sharepoint-auth-bypass-exploitation-2026-08-13.md
  • Created raw/articles/fliff-prediction-market-pivot-2026-08-13.md
  • Created raw/articles/market-snapshot-2026-08-14.md
  • Updated concepts/ai-agents.md, concepts/azure-cloud-operations.md, concepts/cybersecurity-watch.md, concepts/casino-slots-technology.md, concepts/market-watch.md, and index.md

Follow-ups for Sam

  • Check whether any partner, vendor, or legacy environment Casper touches still exposes on-prem SharePoint.
  • Consider a short comparison brief on Microsoft Foundry vs OpenAI vs Google for practical agent economics once one more day of signals accumulates.
  • Watch whether more gaming operators follow Fliff and Novig into prediction-market rails; if yes, this deserves a dedicated comparison page.