Tech News, October 8: Samsung’s $80B AI Windfall, LMCache’s Unpatched RCE, and Claude Enters Google Workspace

Samsung just posted the largest quarterly profit any tech company has ever recorded, powered by AI memory demand. An open-source LLM cache turned up with an unauthenticated remote-code flaw and no patch. A survey of 801 executives found the average enterprise runs 183 AI tools with almost nothing connecting them. And Anthropic moved Claude into Google Docs, Sheets, and Slides. Here is what matters from the last 24 hours.

Samsung forecasts a record $80.2 billion quarter

Samsung Electronics guided on Thursday to an estimated third-quarter operating profit of 107.4 trillion won (about $80.17 billion), up roughly 783 percent from 12.17 trillion won a year earlier, per Reuters. That is the first time any South Korean company has cleared 100 trillion won in a single quarter, and it would be Samsung’s fourth consecutive record quarter.

Revenue is expected up 127 percent to 195 trillion won, also a record. The driver is not phones: it is memory. Tight supply of conventional DRAM and NAND plus surging demand for high-bandwidth memory (HBM) used in AI data centers has pushed chip prices sharply higher, and Samsung is the world’s largest memory maker.

The guidance beat the LSEG SmartEstimate of 106.1 trillion won. Still, analysts flagged that memory price gains are slowing and a stronger won erodes overseas earnings; Reuters notes the demand-supply imbalance is expected to persist into 2027, with shortages into 2028 possible. Samsung releases detailed results, including division breakdowns, on October 29.

LMCache has a 9.8 RCE with no patch available

JFrog disclosed CVE-2026-105192 on October 7: a critical flaw in LMCache, open-source software that caches KV data to speed up LLM servers like vLLM. In multiprocess mode, LMCache runs as a standalone server that LLM workers reach over the ZeroMQ messaging library, and a single crafted network message can execute commands as the user the LMCache process runs as. No login required.

The rating is CVSS 9.8. It affects LMCache from version 0.3.9 (released October 2025) through 0.5.5, the latest stable release, plus the 0.5.6 release candidates and the development branch. As of publication, no fixed version exists.

Exposure depends on one setting. The multiprocess server defaults to localhost, which remote attackers cannot reach. It becomes reachable when an operator binds it to a routable address, as multi-node deployments do, and LMCache’s own example Kubernetes deployment listens on every interface. LMCache running inside a single vLLM process opens no port at all. If you run LMCache multiprocess, check whether the server socket is bound to 0.0.0.0 and restrict it until a patch ships.

Enterprises run 183 AI tools with nothing connecting them

WSJ Intelligence and the agency Code and Theory surveyed 801 C-suite executives at US companies with at least $500 million in annual revenue. The headline number: the average enterprise now runs 183 AI tools, yet 85 percent say they lack the systems to connect them. The report calls this the “orchestration gap.”

The details are grim: 85 percent of the C-suite say AI must now deliver measurable ROI, but three in five say tool proliferation has left strategic execution stagnant, and two-thirds routinely find redundant AI investments. Only 15 percent say their AI investments operate as a unified system with shared intelligence, and more than two-thirds have not even issued an RFP for a centralized orchestration system.

Accountability is the root problem. 42 percent of CEOs believe they own AI orchestration, 57 percent of technology leaders say they do, and one in four companies admit nobody is clearly accountable. The barriers executives cite are cultural and structural, not technical: siloed mindsets (61 percent), fragmented data (58 percent), no central operating layer (56 percent).

Claude for Google Workspace enters public beta

Anthropic announced on October 6 that Claude for Google Workspace is now in public beta for all paid Claude plans. The add-on, installed from the Google Workspace Marketplace, opens Claude in a sidebar next to an open Google Doc, Sheet, or Slide. In Docs it can rewrite selected passages or propose larger edits; in Sheets it writes native formulas, builds pivot tables and charts, and can run heavier data work through a Python environment; in Slides it generates slides matching the deck’s theme and flags layout problems like overlapping text.

Two modes control autonomy: “Ask before edits” (the default) queues each change as an approval card, while “Accept all edits” lets Claude apply changes without stopping. Anthropic also released beta connectors so Claude can create and edit Google files directly from the chat interface.

This puts Claude in direct competition with Gemini inside Workspace and Microsoft 365 Copilot, and the framing is explicit: Anthropic’s sidebar carries the same enterprise controls (Compliance API, customer-managed encryption keys, OpenTelemetry audit export) that IT admins already use. Note that company policies may block it; admins must allow Claude as a trusted app for the add-on to work.

Vivo starts rolling OriginOS 7 and Android 17 to dozens of phones

Vivo has published its rollout schedule for OriginOS 7, which brings Android 17 and an on-device AI agent to roughly 40 devices in India, starting at the end of October 2026 with the X Fold5, X300 Ultra, X300 Pro, X300, and X300 FE, and continuing through February 2027 for older V, S, T, and Y series phones.

Feature-wise, the release leans on practical tooling: a Smart Call Assistant that answers calls you cannot take and summarizes them in the call log, an AI agent that can act on a paired PC, Jovi Security confirmations for sensitive agent actions, Office Kit to use a Vivo phone as a webcam, and cross-brand file transfer that skips mobile data. Notably, Vivo claims cross-device support for Apple Watch, AirPods, and iCloud in supported regions.

The numbers that matter

Key takeaways

• The AI money is in memory, not models. Samsung’s record is a supply story: DRAM, NAND, and HBM demand outstripping supply through at least 2027. Expect memory prices, and device costs, to keep rising.
• Check your LMCache bind address today. Default localhost deployments are safe; anything listening on a routable address with no patch available is exposed to unauthenticated remote code execution.
• AI ROI is an integration problem, not a model problem. 183 tools per company, 85 percent unconnected, and a quarter of firms with nobody accountable explains why budgets are growing while execution stalls.
• Claude now works where your files live. If your team already pays for Claude, the Workspace add-on is worth a trial; keep it in “Ask before edits” until you trust its change queue.
• Android 17 is shipping in volume soon. Vivo’s schedule puts the new OS on dozens of devices starting this month; the cross-device and call-assistant features are the parts worth watching.

References

• Reuters: Samsung Q3 profit jumps nearly nine-fold as AI memory boom lifts chip earnings (Oct 8, 2026)
• The Hacker News: Unpatched Critical LMCache Flaw Lets Unauthenticated Attackers Run Code Remotely (Oct 7, 2026)
• Advanced Television: Report: AI investment not driving enterprise performance (Oct 8, 2026, citing WSJ Intelligence and Code and Theory)
• Gadgets 360: Claude for Google Workspace With Docs, Sheets and Slides Support Launched in Beta for Paid Users (Oct 7, 2026)
• TalkAndroid: Vivo’s OriginOS 7 Brings Android 17 and an AI Agent to 40 Phones (Oct 6, 2026)