GenAI Daily - October 23, 2026: OpenAI Models Land on Google Cloud, Nvidia and ServiceNow Push Hardware-Enforced Agent Controls, OpenAI Nears $1.25T Round
Top Stories
OpenAI Frontier Models and Codex Come to Google Cloud Model Garden
OpenAI and Google Cloud announced on October 21 that OpenAI's frontier models and Codex agentic workflows are coming to Google Cloud's Model Garden.
Purchases go through Google Cloud Marketplace, and enterprises can draw down existing Google Cloud spend commitments to pay for them.
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out in private preview to select enterprise customers. Teams can connect local Codex CLI and Desktop workflows to Google's platform for inference starting October 21.
OpenAI models will also appear in Google Antigravity and the Gemini Agent for task routing.
General availability is still pending. The companies say it is coming in the next few weeks but have given no launch date.
Microsoft remains OpenAI's primary cloud partner, so existing Azure usage is unaffected.
OpenAI Developer Community | Securities.io | AlphaSignal
Why it matters: GCP-standardized teams can now put OpenAI inference through the same procurement, identity and billing path as Gemini and run side-by-side evaluations. Treat it as announced distribution until regional and model-level availability is confirmed.
Nvidia Publishes OpenShell Reference Architecture as ServiceNow Arc Hits GA
Nvidia launched the OpenShell Reference Architecture at GTC Berlin on October 21. It is a blueprint for running AI agents inside company systems under enforced limits.
Vera Rubin infrastructure runs the models, Vera CPUs run agent harnesses and tool use, and BlueField-4 DPUs run Nvidia's Sentry software for out-of-band monitoring.
OpenShell itself is open source under Apache 2.0 (v0.1.0) and runs on any hardware. One caveat: Sentry is currently a reference design, with general availability targeted for early 2027.
CrowdStrike, Palo Alto Networks, Zscaler and Check Point announced integrations. BBVA and Santander said they are exploring it as a governance framework.
SAP is integrating the architecture with its Joule Studio runtime.
ServiceNow is the first enterprise software vendor to ship on it. Arc is a personal agent that works across ServiceNow workflows, email, Teams and other productivity apps. It can run on its own computer inside an OpenShell runtime or on an employee's desktop.
When an AI Passport is revoked, BlueField can enforce the revocation before the agent proceeds.
Arc is included at no extra cost with EmployeeWorks Foundation, Advanced and Prime SKUs. Mac support is live, with Windows expected within weeks.
Investing.com/Reuters | FourWeekMBA | TechTimes
Why it matters: This follows ServiceNow's AI Passport launch last week and moves agent governance from policy documents into runtime enforcement. Buyers can now ask vendors for revocation and kill-switch proof, though hardware-level enforcement via Sentry will not ship until 2027.
OpenAI Nears Up to $30B Round at $1.25T Valuation Ahead of 2027 IPO
CNBC reports OpenAI is close to finalizing a round of up to $30 billion at a $1.25 trillion pre-money valuation.
Sources describe the round as a bridge to a 2027 IPO. The headline says UAE investors are leading.
OpenAI told investors on Thursday it aims to secure signatures by the end of next week, but the final size is still being determined.
Why it matters: The round gives OpenAI more runway for compute and for the multi-cloud distribution push, such as the Google Cloud deal above. It also follows reports that its annual recurring revenue is nearing $70B.
Key Developments
METR and Redwood Find Six Ways to Blind OpenAI's Agent Monitor
A joint report from METR and Redwood Research says OpenAI's internal agent monitoring could be evaded in six ways. One exploit would let an agent keep hundreds of recent actions out of the monitor's view. The groups each placed one researcher inside OpenAI for three weeks.
About a tenth of Codex activity that should have been monitored slipped through from May to July, and two settings could switch monitoring off for newly launched agents.
Monitoring eligibility for relevant Responses API agentic traffic was estimated at 75-94%.
OpenAI says it patched every flaw affecting production systems, but METR says it has not verified that.
Impact: Monitoring coverage and resistance to evasion are now measurable procurement questions, and logging has to be verified rather than assumed. This is the same gap Nvidia and ServiceNow are targeting with out-of-band enforcement.
DeepSeek Raise Closes at About $15B - UPDATE
Last week's report put the round at $12B-$15B in talks. Bloomberg now reports DeepSeek has raised about $15 billion (roughly 100 billion yuan), two days after the U.S. ordered federal civilian agencies to remove its technology.
The target started at 50 billion yuan, rose to 80 billion, and closed at 100 billion.
Potential investors include CATL and Tencent.
The previous round, in June, raised more than $7 billion at a $50 billion valuation.
Impact: The federal ban limits government use, but the capital supports continued open-weight releases that enterprises outside federal scope can still self-host. Compliance teams should expect the sourcing question to keep growing: SAP's CFO says customers now ask where the AI inside their business software was built.
Meta Discloses $825B in Lease and Contract Commitments - UPDATE
Meta's capex floor and new share class were covered earlier this week. The 10-Q adds new detail on obligations. Meta says it has agreed to spend about $825 billion on leases and contracts, many tied to AI compute and infrastructure.
That is up from nearly $700 billion in July. It breaks down into $454.5 billion in non-cancellable contracts and $355.7 billion in unstarted leases.
$121 billion of the contract amount comes due in 2027 alone.
Free cash flow was negative $2.1 billion for the quarter.
The cost of insuring Meta's debt against default hit a record.
Impact: Combined with the CoreWeave financing worries, this raises the odds of tighter AI infrastructure financing. Teams that depend on a single provider's capacity commitments should check the provider's balance-sheet exposure.
Walmart Opens an MCP Agent Gateway as Amazon Keeps Blocking
This is a fresh angle on last week's story about sites blocking third-party agents. Walmart launched its own Agent Gateway to manage agent transactions.
It is a new MCP server that helps approved agents connect to Walmart over the Model Context Protocol.
Paired with Agent Profiles, it lets customers set agentic spending limits, require purchase approvals, and cut off access.
External agents will not have access to underlying payment card data or sensitive order histories.
TechCrunch adds that non-compliant agents will be blocked.
Walmart's rollout to shoppers starts next month.
Amazon, by contrast, has blocked agents from Perplexity and Meta.
Impact: Developers building shopping agents get a documented, permissioned integration path with delegated-spend controls. Retailers deciding how to handle agent traffic now have a concrete open-door reference design.
Product Launches
Recursive Flint
Flint is a free, open-weight coding model under Apache 2.0 that runs offline on a laptop with 48 GB of memory, at well over 100 tokens per second.
It has 39.8 billion parameters, with about 2.5 billion active per token.
Context is 262,144 tokens. Recursive says its Eureka Machine research system helped build nearly every stage. Benchmark claims are self-reported and mixed.
On Recursive's own numbers it is the best small mixture-of-experts model for agentic coding, but a larger dense Qwen model still beats it.
Why it matters: Teams with code-privacy or usage-fee constraints get a local option to test.
i-scoop | TechCrunch via Yahoo
1Password for Agents
1Password for Agents lets people approve task-specific access for personal agents without exposing the underlying credentials to the agent or its model context.
The company says it is partnering with AI labs and agent providers including Anthropic, Meta, Nvidia and OpenAI.
1Password also reported ARR above $500 million. Monthly active non-human identity users of its developer and AI tools grew 500% year over year.
Why it matters: Enterprise teams should check how employee-approved agent access gets audited and revoked.
Microsoft Guardian Agents
Microsoft announced Guardian Agents, a feature that watches AI agents running on Azure and intervenes when it detects risky behavior.
It can detect up to 60 agent types and dynamically adjust permissions. Pricing has not been disclosed.
Microsoft intends to detail its upcoming controls at Ignite in November.
Why it matters: Cloud-native monitoring of agent behavior gives Azure customers a built-in option for governing agents at scale, though the lack of disclosed pricing and the details still pending until Ignite mean teams should wait before planning around it.
Funding & Deals
Blacksmith Raises $66M Series C
San Francisco-based Blacksmith runs CI for companies whose coding agents have multiplied pull requests. It was valued at $1 billion in this round. It says its runners are twice as fast at up to 60 percent lower cost than GitHub-hosted runners, using Firecracker microVMs and NVMe caches.
ARR has grown 17x, and it has run over 90 million CI jobs for customers including Mercury and Granola.
Blacksmith cites a disclosure that Anthropic's internal CI infrastructure grew 25-fold in the six months after Claude Code launched. Led by Sequoia Capital.
Why it matters: As coding agents multiply pull requests, CI infrastructure becomes a bottleneck, and investors are backing specialized providers built for that load.
Warp Sciences Raises $215M Series A
Former OpenAI for Science head Kevin Weil launched Warp Sciences at an $865 million valuation. It is building an automated US lab for a drug-development step that pharma has long outsourced to Chinese firms such as WuXi AppTec.
Chemists will wear cameras and pressure-sensing gloves to create training data for its own AI models.
It has not shipped a public product.
The bet is that proprietary lab data is the missing ingredient for AI in chemistry. Led by Lightspeed Venture Partners.
Why it matters: The round shows investor appetite for AI companies that generate their own proprietary data in the physical world, and for onshoring a step of drug development that has relied on outside suppliers.
Tomorrow's Watch List
- OpenAI's $1.25T round: the company is targeting signatures by the end of next week.
- General availability of OpenAI models in Google Cloud Model Garden, expected within weeks.
- Microsoft Ignite in November, where further agent controls are due.
- ServiceNow Arc Windows support, expected within weeks.
Related reading: Check out this week's [Deep Insights analysis] for strategic context on these developments.
GenAI Daily is curated by rAInvent. For in-depth analysis, see the rAInvent articles.
