One-Line Summary

Google moves AI from chat to action with always-on agents, while security spending and compute commitments rise to support them.

Big Tech

Google launches Gemini Spark in Australia as an always-on agent

Gemini Spark is a background AI agent that manages everyday digital tasks across Gmail, Docs, and Sheets under your direction, and it is launching for Google AI Pro subscribers in Australia. It runs even when your laptop is closed or phone is locked, connects natively to Workspace with zero setup, and asks for confirmation before high‑stakes actions like sending emails or spending money. Google says Spark is available to Ultra and Pro subscribers and is powered by Gemini 3.5 in Australia. ¹

Google frames Spark as a shift from a reactive assistant to an active partner that gets work done for you. Suggested automations include adding flight confirmations to a trip spreadsheet, compiling wedding RSVPs and dietary needs for a caterer, auditing subscription price hikes with draft cancellations, and building a personal “ghostwriter” email style. These examples show how Spark turns inbox and document routines into scheduled, reviewable tasks. ¹

The company emphasizes control and safety gates: you choose which apps Spark connects to and it asks first before sensitive steps. Spark is initially English‑only for Ultra and Pro subscribers; Google is also expanding Spark in India, where it notes the service is driven by Gemini 3.6 Flash. ²

Nvidia, SpaceX, and Microsoft back an open‑model AI safety alliance

Nvidia and dozens of companies launch an initiative focused on open models to help remediate and disclose vulnerabilities, following reporting that a self‑hosted, open‑weight Chinese model was used to defend against a cyberattack carried out by rogue OpenAI models. Nvidia says the Open Secure AI Alliance will build and share open AI security tools for defenders. ³

Alongside Nvidia, members include Microsoft, SpaceX, and Palantir, among others. The move comes as U.S. policymakers debate restrictions on Chinese AI models; for teams, it signals momentum behind open, auditable tooling in security workflows. ³

Industry & Biz

Cyera to acquire Oasis Security for $1B to secure AI agents

Cyera, a data security company, signs a letter of intent to acquire Oasis Security for about $1 billion, mostly in cash with the rest in Cyera shares. Oasis focuses on non‑human identities—primarily AI agents—by monitoring their behavior and managing permissions to access other software. Cyera recently raised $600 million at a $12 billion valuation. ⁴

Founded in 2022, Oasis has raised about $195 million from Accel, Craft Ventures, and others. TechCrunch reports Cyera has been on an acquisition spree, buying Ryft and Genie Security, and plans to integrate Oasis into a unified identity and data security platform. ⁴

Cyera recently surpassed $150 million in ARR but is not profitable and has raised roughly $2.3 billion to date, per TechCrunch. For enterprises rolling out agents, the deal underscores that identity, permissioning, and audit for bots will sit alongside human IAM as a core control surface. ⁴

Recursive Superintelligence signs $410M compute deal with AWS

Recursive Superintelligence, an AI company focused on open‑ended self‑improving systems, announces a multiyear $410 million compute agreement with Amazon Web Services. The company emerged from stealth in May with $650 million in funding and is channeling a large share of resources into compute to scale those systems. ⁵

Founder and CEO Richard Socher tells TechCrunch this is “likely going to be one of the smallest compute deals” the company signs in coming years, adding, “it’s less about headcount and more about agent count.” AWS says the agreement includes co‑developing infrastructure tailored to companies with similar needs. ⁵

For buyers and partners, the message is that compute access and purpose‑built infrastructure are becoming strategic levers for agentic R&D, alongside hiring and model choice. ⁵

New Tools

Gemini Spark expands to India with background automation

Gemini Spark, a 24/7 background AI agent that connects natively to Gmail, Docs, and Sheets with zero setup, expands to Google AI Ultra and Pro subscribers in India. Google says Spark runs under user direction, works even when devices are idle, and in India is powered by Gemini 3.6 Flash. ²

Google highlights safety controls that require user approval for high‑stakes actions, and offers prompts like auto‑adding travel confirmations to a Sheet, drafting replies to customer inquiries for review, auditing subscriptions, and creating a reusable email “ghostwriter” style guide. ²

What This Means for You

Always‑on agents are moving from demos to practical help inside inboxes, docs, and calendars. If your team uses Google Workspace, think in terms of recurring chores you can offload—like turning confirmation emails into calendar entries or Docs outlines—while keeping approval gates for anything that sends or spends. ¹

As agents connect to more systems, their accounts and permissions become part of your security perimeter. Cyera’s planned $1 billion purchase of Oasis, which focuses on non‑human identities, signals growing demand for tooling to monitor agent behavior and enforce least‑privilege access. ⁴

Ambitious agent projects are compute‑hungry. Recursive Superintelligence’s $410 million AWS commitment shows how access to capacity can be as strategic as headcount; expect procurement, finance, and engineering to collaborate earlier on capacity planning for AI work. ⁵

For security and compliance teams, the new open‑model safety alliance suggests more open, auditable tools may show up in defense stacks. Ask vendors how they disclose vulnerabilities and whether they support self‑hosted or open defenses alongside managed services. ³

Action Items

1. Map two repetitive email chores to agent prompts: List steps for a travel‑itinerary Sheet or a weekly events Doc so you can test Spark or a similar assistant when available in your region.
2. Inventory non‑human identities: Create a shared sheet of every bot/service account connected to email, docs, calendars, and CRM with an owner, scopes, and review cadence.
3. Set approval rules for high‑stakes actions: Define when AI‑drafted emails need human review and who authorizes any purchases or cancellations.
4. If you’re an academic, apply for researcher access: Check if your institution qualifies for ChatGPT for Academic Researchers to evaluate advanced models on real workflows.