Google Cloud Blog

[Google Cloud] Authentic Brands Group, the global lifestyle and entertainment platform whose portfolio spans more than 50 brands including Reebok and Juicy Couture, has built AuthOS, an AI operating system powered by Gemini Enterprise Agent Platform and Cloud Run, according to Google Cloud. The system's 50 interconnected AI agents and tools are designed to help teams create content faster, streamline workflows and reduce time spent across disconnected tools. According to Google Cloud, more than 80% of employees use AuthOS weekly, and AI-enhanced product images for Reebok delivered more than 60% higher return on ad spend compared with unenhanced imagery. AuthOS uses Vertex AI's modular architecture to incorporate models from Model Garden, Google Cloud said.

[Google Cloud] Google Cloud described its experiment with ClusterNetworkPolicy (CNP) in Google Kubernetes Engine (GKE), saying standard Network Policies limit policy application to the namespace level while CNP lets administrators apply granular networking policies across an entire cluster and scale security and governance policies. According to Google Cloud, CNP follows a three-tier precedence flow: an admin tier at top precedence, where cluster administrators enforce mandatory security rules that cannot be bypassed, such as blocking metadata access; a network policy tier in the middle, for namespace-scoped policies managed by application teams to control ingress and egress for their microservices; and a baseline tier at bottom precedence, where administrators set fallback defaults such as a default-deny posture across namespaces. The Codelab experiment aims to restrict traffic to the LLM namespace, allow only authorized clients and prevent cluster-wide access to the metadata API except for DNS port 53. It uses a custom VPC with Cloud NAT, a GKE Standard cluster running GKE Dataplane V2, a TPU v5e node pool in a 2x2 topology, and Gemma 4 served with vLLM on port 8000, with three namespaces: gemma-llm, client-ns and untrusted-tenant.

[Google Cloud] Google Cloud said on November 10, 2026 that it launched the Gemini factory, an intensive co-innovation initiative designed to help independent software vendors take agentic Gemini-based solutions from a blank whiteboard to listings on Google Cloud Marketplace. According to Google Cloud, specialists work side-by-side with partner teams through hands-on workshops in global technology hubs including San Francisco, New York City, Sunnyvale, London, Paris, Tel Aviv and Tokyo to ideate, architect and build enterprise AI solutions. Google Cloud said the program helps partners choose among MCP connectors, skills, plugins and autonomous agents, compress integration work from months to days, and accelerate customer activation through funding incentives and go-to-market support.

[Google Cloud] Google Cloud released a reference architecture that wires automated CodeMender remediation directly into Git pull requests via Cloud Run. According to the Google Cloud blog, the agent returns a confidence score on a vulnerability finding during code analysis; if the score is below 100%, cm verify can build and simulate an exploit in a sandbox to determine whether the vulnerability is actually exploitable, and the simulation results can change the severity. Because the verification sandbox is ephemeral, the pipeline automatically streams these artifacts to Google Cloud Storage to preserve an audit trail. Google Cloud said developers can reply /verify and /fix to see exploit traces and code diffs within the pull request timeline, and after /approve the agent commits the patch to the branch, pushes it and sets the security/codemender-gate commit status to green; branch protection is what turns that status into a merge block, and without it the gate is advisory. Google Cloud also cautioned that verification pipelines must separate true negative results from test crashes, flagging an exploit that fails to complete cleanly as inconclusive rather than safe.

[Google Cloud] Nearly half (49%) of 1,250 U.S. enterprise IT and business decision-makers surveyed by Google Cloud and Hypothesis Group said reliance on on-premises software has actively delayed, reduced in scope, paused or canceled AI or other technology initiatives, according to the survey. The study defined legacy technology as software running locally on computers and servers rather than natively in the cloud. 39% of decision-makers said reliance on legacy software and cloud providers forces them to navigate complex licensing rules or contracts, or limited negotiating flexibility. 80% of leaders agreed that a provider's installed base should not determine which AI technology an organization uses, while 78% said the providers that led the previous era of enterprise technology may not be best positioned to lead the AI era. Google Cloud said nearly 90% of the Fortune 100 are already using Gemini Enterprise to build, scale and govern agents.

[Google Cloud] Google Cloud said at KubeCon + CloudNativeCon North America 2026 that it is introducing and expanding a unified, open-source AI-native stack across Kubernetes and Google Kubernetes Engine (GKE). According to Google Cloud, the stack includes OpenRL, a Kubernetes-native post-training service; llm-d 1.0, an open distributed inference control plane with prefix-cache-aware routing, disaggregated serving and multi-tier KV cache management; Ray on GKE with KubeRay; Agent Substrate, an orchestration runtime for multi-turn AI agent workflows; gVisor, a kernel-level container sandbox; and TorchTPU, a backend providing native PyTorch support across Cloud TPUs and GPUs inside upstream vLLM and SGLang. Google Cloud said all of these components are CNCF Sandbox projects. The company said the stack is aimed at challenges including accelerator underutilization in RL post-training, cache-blind load balancing, framework fragmentation and agent security in multi-tenant clusters, and that it is offering sample manifests for users to try on their own Kubernetes clusters.

[Google Cloud] Google Cloud announced on November 10 the launch of Gemini for healthcare, describing it as a purpose-built industry solution that extends the universal Gemini agent introduced last month. According to Google Cloud, healthcare data remains fragmented across disconnected EHRs, claims systems and administrative tools, and clinicians spend nearly 28 hours a week manually reconciling information. The offering has four core components: healthcare skills written by domain experts, connectors to EHR, claims and administrative systems, an open healthcare partner ecosystem, and a secure governed platform with open model choice. Google Cloud said the product is built to support HIPAA compliance and is available in preview starting today.

[Google Cloud] Data Cloud connectors in Gemini for business are now in preview, giving business users and analysts a direct way to query live enterprise data in natural language, according to Google Cloud. The connectors work with Knowledge Catalog to locate relevant data, interpret the question and execute permission-scoped queries under the user's identity, enforcing row- and column-level security and respecting end-user credentials. Users can pull data from BigQuery, operational databases including Spanner, AlloyDB and Cloud SQL, and governed data from Looker, then move findings into tools such as Google Docs, Gmail drafts or Google Slides. According to Google Cloud, the BigQuery connector also supports querying open table formats including Apache Iceberg, Delta Lake and Parquet, as well as cross-cloud datasets in AWS or Azure.

[Google Cloud] Google Cloud said its Cloud Pathways distributed machine-learning runtime for massive accelerator fleets is now generally available on Google Kubernetes Engine (GKE), positioning it as a core runtime pillar of the Google Cloud AI Hypercomputer. Pathways was originally engineered by Google DeepMind to train and serve Google's frontier models, including Gemini. According to Google Cloud, leading AI companies and global enterprises processed more than 130 million Cloud TPU chip-hours using Pathways on GKE over the last 90 days, across a broad range of TPU generations. Cloud Pathways uses a single-controller architecture built on the open-source Interim Framework Runtime (IFRT) and offers colocated Python sidecars, elastic fault-tolerant training, an interactive mode and a Shared Pathways Service. Google Cloud also announced a Preview integration with Cluster Director to extend Cloud Pathways to Slurm clusters, supporting unified multi-slice JAX execution, elastic training and interactive notebooks.

[Google Cloud] Google Cloud CISO Chris Betz published the company's 2027 cybersecurity forecast on November 9, 2026, drawing on insights from leaders at Google, Google Cloud and Mandiant. Betz said that at the start of the year only a few organizations used LLMs and agents to find and fix vulnerabilities, while today it is happening at scale across enterprises, and that agentic hacking's scale, speed and sophistication are outpacing human-only response. Karthik Narain, chief product and customer officer at Google Cloud, said the risks in 2027 will not be just server outages or stolen databases, and that the data and technology foundation agents rely on also matters; every consequential action by an autonomous system requires a complete, tamper-proof decision trace recording inputs, how the model reasoned, which tools it called, what parameters it passed and why policy allowed it, so teams can reconstruct, validate and audit automated decisions before scaling. Ramya Chitrakar, vice-president of Engineering at Google Cloud, said 2027 is the year analysts stop chasing alerts and start shaping outcomes.

[Google Cloud] Google Cloud said on November 9, 2026 that it is expanding its eighth-generation TPU portfolio with two additional systems: TPU 8i Max, aimed at massive context windows and the highest compute density per rack, and TPU 8tx, built for secure and confidential inference. According to Google Cloud, this is the first time it has introduced four TPU systems in a single year, following the April Google Cloud Next launch of TPU 8t for large-scale training and TPU 8i for real-time reasoning; the company said those two offer up to 2.7 times better training performance and 80% better inference price performance, respectively, versus its previous generation. Google Cloud said TPUs are now a rapidly growing business with nearly 100 customers, including frontier labs, trading firms, retailers and universities, and that customers can now deploy TPUs directly into their own data centers.

[Google Cloud] Google Cloud said GKE dynamic slicing is now generally available, decoupling logical TPU slice formation from physical nodepool provisioning. According to Google Cloud, dynamic slicing reduces AI workload startup times by up to 5x and workload recovery by 4.5x compared with creating a static nodepool for each TPU slice. The feature uses TPU optical circuit switching to stitch together TPU cubes, supporting super-slicing beyond a single cube and sub-slicing a cube into smaller slices for multiple workloads within one nodepool. Google Cloud said the approach eliminates the delete-and-create nodepool cycle, limits large slice failures to a 16-VM 4x4x4 slice topology, swaps unhealthy cubes in one to two minutes and allows use of partially healthy nodepools.

[Google Cloud] Google Cloud introduced TPU Sync, an open-source native C++ library for moving model state across TPU clusters. According to Google Cloud, the library treats accelerator HBM, host DRAM, shared storage and the data center network as a single co-designed memory hierarchy, using asynchronous DMA transfers to overlap data movement with active computation. For inference, Google Cloud said TPU Sync turns host DRAM, local SSDs and Google Cloud Managed Lustre into an automated multi-tier KV cache; for reinforcement learning, it exposes low-level DMA and DCN primitives that higher-level frameworks can use to stream updated weights from trainers to samplers. The company said TPU Sync cuts KV transfer latency by more than 90% and reduces Time to First Token by an average of 90% for heavyweight open-weight MoE models of 235B to 480B parameters in prefix-caching and disaggregated-serving workloads, compared with running without it. TPU Sync integrates with PyTorch via TorchTPU and with JAX, and serves as the TPU KV cache connector for vLLM, SGLang and llm-d disaggregated serving, according to Google Cloud.

[Google Cloud] Google Cloud said its new .bigquery accessor from BigQuery DataFrames (BigFrames) lets users run machine learning workflows, including forecasting, classification and high-dimensional embedding generation, without leaving a pandas DataFrame. According to Google Cloud, the accessor translates pandas operations into Google Standard SQL, executes foundation models such as TabFM natively on BigQuery infrastructure and returns results as a standard pandas DataFrame. The company said ai.predict() can perform regression directly over two pandas dataframes with a single call, bypassing local scikit-learn training. Google Cloud said users need BigQuery Job User and Agent Platform User roles, and must install pandas and bigframes in a Jupyter environment and specify a Google Cloud project ID.

[Google Cloud] Mito used Google Cloud's serverless, asynchronous architecture to produce a 120-second World Cup TV spot for Movistar Plus+, part of Telefónica, according to a Google Cloud blog post. Mito said the project required generating more than 5,000 assets, including 1,000 videos, across roughly 65 iteration rounds, and that the ad aired on every major television channel in Spain; the production team described it as 95% AI-generated. Mito runs its API, AI Director and background workers as Java services on Cloud Run, decouples job execution with Pub/Sub, stores assets in Cloud Storage and delivers them via Cloud CDN, and accesses models through Vertex AI, according to the post.

[Google Cloud] Google Cloud said its next-generation storage-optimized Z4M machine series is now generally available in Google Compute Engine as both virtual machine and bare-metal instances. According to Google Cloud, Z4M targets AI/ML training and inference workloads such as checkpointing, distributed key-value cache and parallel file systems. The instances are powered by fifth-generation Intel Xeon Scalable processors and offer up to 168,000 GiB of local SSD storage with up to 400 Gbps of Cloud RDMA networking; RDMA reduces data-access latency by up to 60% compared with TCP networking, Google Cloud said. Z4M VMs provide 875 GiB of local SSD per vCPU, with up to 192 vCPUs and up to 1,536 GiB of memory. Google Cloud said Z4M delivers up to 15.6M random-read IOPS and up to 75,600 MiB/s sequential-read throughput, improvements of up to 70% and 110% respectively over Z3.

[Google Cloud] Gemini Omni 1.2 Flash is now generally available on Google Cloud, the company said, targeting the everyday videos businesses produce in volume, such as product videos, virtual try-ons and ads. According to Google Cloud, the model is priced at $0.07 per second of 720p video, 30% less than Gemini Omni 1.1 Flash, with similar savings at other resolutions, and can generate and edit clips of up to 15 seconds, up from 10 seconds. Google Cloud said the model follows image and video references more closely, keeping products, characters and brand details consistent from shot to shot; users can draft at 360p or 720p and upscale selected results to 1080p or 4K, and refine the same clip in conversation by swapping a product, changing lighting or changing camera angle. Audio references are gated behind allowlisting. Google Cloud said the model is production-ready, backed by its enterprise-grade infrastructure and security, and that outputs carry an imperceptible SynthID watermark. Customers including Artlist, FLORA, invideo and Krea said they use it for reference-based generation, video extension and style transfer workflows.

[Google Cloud] Google Cloud said BigQuery has introduced AI.AGG, a preview multi-level semantic aggregation function embedded natively in GoogleSQL, which treats semantic aggregation as a distributed map-reduce problem and lets data teams run natural-language instructions over millions of rows with a single SQL query. According to Google Cloud, the engine automatically performs row-level validation, dropping or returning an error on payloads above 10 MiB; groups input rows into context-safe batches that fit the token limits of the chosen Gemini endpoint; applies the prompt to each batch in parallel via Vertex AI to generate intermediate summaries; and recursively batches and re-submits those outputs until a single aggregated string is produced per SQL GROUP BY clause. Google Cloud said AI.AGG relies entirely on full-weight LLMs at every pipeline stage, unlike row-level functions such as AI.CLASSIFY that can use distilled CPU proxy models. The function natively supports text and multimodal inputs, and with BigQuery Object Tables and OBJ.GET_ACCESS_URL, teams can aggregate visual insights from images in Cloud Storage without data movement or transcoding, according to the company. Google Cloud cautioned that when a STRUCT is passed, a single NULL field causes the entire row to be silently discarded, and recommended wrapping nullable fields with IFNULL().

[Google Cloud] Google Cloud said its team built an agent skill optimizer using AlphaEvolve, Harbor and Gemini, treating the text of SKILL.md skills as an evolvable program in which Gemini proposes new versions, an evaluator scores them and the search keeps the higher-scoring version. According to Google Cloud, on the public Terminal-Bench 2.0 benchmark with OpenCode running Gemini 3.8 Flash, the agent passed about two-thirds of its attempts with the baseline skill, but results were about the same without the skill; after a few AlphaEvolve iterations, the team saw improvements of 5%. Google Cloud said Harbor runs real coding agents on Google Kubernetes Engine with one pod per trial, and a single search ran thousands of agent trials, with the evaluator returning pass and failure scores plus short notes on failed or timed-out tasks that Gemini uses for the next version. Google Cloud also said that in an early run on the same benchmark its model's latency explained much of the variance in candidate scores, so each candidate runs next to the baseline skill on the same tasks at the same time and scores are the difference; about a third of the search's trials hit their time limit, and the best skill is built around that limit.

[Google Cloud] More than 2,700 government leaders, innovators and industry partners convened over two days at the Google Public Sector Summit in Washington, D.C., under the theme "Where mission meets moonshot," according to Google Cloud. In the morning keynote, Karen Dahut, CEO of Google Public Sector, said organizations including NASA, NOAA, the Department of War, the Department of Energy, the State of Connecticut, the U.S. Army and Case Western Reserve University are collaborating with Google to drive impact in the agentic era. Thomas Kurian, CEO of Google Cloud, reinforced the company's commitment and shared innovations including Gemini for government services, the Wiz Red, Green and Blue agents now available within Wiz for Gov, Wiz's FedRAMP Class D (High) certified environment, and enterprise search capabilities for Gemini on Google Distributed Cloud for air-gapped deployments.

[Google Cloud] Google Cloud announced security updates for the agentic era at its Security Summit. According to Google Cloud, its autonomous code-security agent CodeMender helps organizations find and fix code vulnerabilities at machine speed and can eliminate manual triage bottlenecks. CodeMender works with multiple models, including Gemini 4 Argon, and can identify previously unknown vulnerabilities and generate tested patches before they can be exploited. According to Google Cloud, Wiz recently used Gemini 4 Argon and other models to identify and help remediate core PHP vulnerabilities affecting 70% of websites, with Gemini 4 Argon contributing to finding the vulnerability, developing a likely exploit path, detecting attacks and validating the patch. Currently in preview, CodeMender adds secure isolation and flexible scanning: it isolates vulnerability verification and exploit simulation inside an operating system-level sandbox or isolated virtual machine and adds human-in-the-loop checks, and it can scan a specific folder, an entire repository or only files modified since the last scan. Google Cloud said hundreds of customers have used CodeMender since the preview became available in July. Ivan Fontarensky, CTO of Cybersecurity Services at Thales Group, said early evaluation shows the potential to reduce vulnerability discovery time by over 95% while keeping expert validation and remediation under human oversight.

[Google Cloud] Mercari, a consumer-to-consumer marketplace with more than 20 million monthly active users, selected Chrome Enterprise Premium and Chromebooks to secure its customer service environment, according to a Google Cloud blog post by Yu Susaki, manager of enterprise security at Mercari. Mercari said its prior setup relied on legacy devices routed through virtual desktop infrastructure (VDI), creating operational friction, a poor user experience, higher IT overhead and overlapping data-loss-prevention controls. The company shifted from OS-level isolation to browser-layer data control, configuring strict DLP rules that restrict users from pasting or uploading files to destinations outside approved business tools. Mercari said the transition was smooth because employees already used the Chrome Enterprise browser, resulting in strong adoption. With zero-touch enrollment, reduced VDI dependency and consolidation of security agents including legacy DLP, MDM and EDR, Mercari projects a cost reduction of more than 50% over three years in this environment, covering device costs, VDI running costs and security licenses. Mercari also said it is using browser policies to enable a safe, read-only adoption of Gemini in Chrome.

[Google Cloud] Google Cloud said API multi-versioning for Compute Engine is now generally available, letting customers adopt modern API semantics while preserving the stability of the /v1 API. According to Google Cloud, stable versions use YYYY-MM-DD identifiers and ship on a quarterly cadence, starting with 2026-09-01; each is a locked, point-in-time contract that receives no breaking changes after publication and has no end-of-support date today. Preview versions, named YYYY-MM-DD-preview, branch from a stable parent and carry a one-year lifespan. The /v1 API remains fully supported and is the default when no date-based version is specified, so existing applications continue to work without modification, Google Cloud said. Users select a version via the X-Goog-Api-Version HTTP header or the $apiVersion query parameter, while resource URIs keep the /compute/v1/ base path.

[Google Cloud] Google's AI & Infrastructure team said on Nov. 3 that it has launched Physical-Engineering's Last Exam (PLE), a benchmark and open competition developed with Mercor to test frontier models on mechanical, electrical, civil and structural engineering tasks. According to Google Cloud, it released 29 public tasks, grading tools and a leaderboard. PLE comprises 85 long-horizon tasks: 30 mechanical, 27 electrical, 18 civil and 10 structural, with each task averaging 61 files and 5,170 files in total, including PDFs, spreadsheets, emails, 2D CAD drawings, 3D BIM models and document revisions. Google Cloud said 44 tasks were expert-authored by practicing engineers, while 41 were designed by engineers and assembled by an agent pipeline, then validated and made harder by engineers. The team tested 14 frontier models using seven public agent harnesses. According to Google Cloud, all frontier models scored below 50% on the private test set, with the top model on its native harness reaching a mean score of 48.3%; on co-authored tasks, the highest-scoring agent configuration satisfied 91% of individual constraints, but only 41.3% of its deliverables passed all hard safety checks.

[Google Cloud] Google Cloud said its team has spent recent weeks measuring and optimizing its Google Cloud Developer Plugin, running dozens of scenarios with Gemini, Antigravity and other agents, using Claude Sonnet 5 on the Gemini platform inside isolated Harbor pods on GKE Autopilot, and contributing Kubernetes and Gemini platform capabilities upstream to NVIDIA's open-source evaluator. The team outlined five engineering rules for verifying whether a skill is worth its tokens. It said trigger collisions and security risks should be gated before containers are spun up, because coding agents do not read full SKILL.md bodies at turn zero and the router only inspects the short name and description in the YAML frontmatter. It also warned against trusting evaluations run on a developer laptop, where cached credentials and exported project variables contaminate the baseline, and said teams should measure paired WITH versus W/OUT ablation at Pass@3 rather than rely on impressions. According to the team, more than a third of benchmark tasks in its retrieval study produced different scores across consecutive runs of the exact same code and prompt.

[Google Cloud] Google Cloud Consulting said in a Nov. 3 post that as enterprises move from generative AI that answers questions to agentic systems that carry out multi-step plans, budget, compute and tooling can help launch a program but adoption depends on intentional culture work. Based on its work with companies, the team offered five lessons: make workflows agent-executed and human-directed, shifting from step-by-step human validation to human-in-the-lead models with parallel dependencies and exception-based management; turn hesitation into ownership by delegating to agents like junior team members, setting clear decision boundaries and starting with low-stakes use cases; manage hybrid teams of people and agents by focusing on results and rewarding those who guide and audit agents; and build agentic literacy through experimentation so employees can judge when an agent's logic drifts from company goals.

[Google Cloud] Fintech firm Nammu21 partnered with Google Cloud on a proof-of-concept that uses its AI-powered platform to structure legal agreements and instantiate them as smart contracts on the Google Cloud Universal Ledger (GCUL), automating the entire loan lifecycle. According to the Loan Market Association, secondary market settlement times for par/near par loans in Q4 2024 still averaged 42 days, with a median of 27 days. Nammu21 renders each facility's governing terms as a Digital Credit Instrument (DCI), mapping every term back to its originating legal provision and expressing it as executable behaviour against a loan's live state. GCUL is an institutional-grade, permissioned distributed ledger that delivers near-instant transaction finality; its contracts are written in gculpy, a strict, statically typed subset of Python, and the ledger requires explicit user authorization before a contract can store data on their behalf.

[Google Cloud] Google Cloud said it is previewing direct Iceberg access in AlloyDB for PostgreSQL, letting developers query Apache Iceberg tables stored in a cloud storage bucket directly from AlloyDB and join them with live operational data in a single SQL query. The company also announced general availability of AlloyDB sync and Lakehouse federation. According to Google Cloud, the three options cover different data locations and read patterns: direct Iceberg access reads tables in place, accelerated by the AlloyDB columnar engine; AlloyDB sync serves lakehouse data from AlloyDB for latency-sensitive workloads on a user-set schedule; and Lakehouse federation queries native BigQuery tables in place with BigQuery doing the processing. Google Cloud said that in its testing on a 1 TB dataset in Cloud Storage, AlloyDB direct Iceberg access was up to 7x faster on aggregation queries than Amazon's comparable PostgreSQL-compatible service and up to 3.5x faster than the open-source PostgreSQL extension.

[Google Cloud] Google Cloud on Nov. 4 announced updates to its Agentic Data Cloud. According to Google Cloud, its borderless Lakehouse, built on the open Apache Iceberg format, gives enterprises one governed foundation for data across multi-cloud, on-premises and SaaS platforms, and Apache Iceberg data under management has more than tripled over the past year. New capabilities include Gemini working directly on the borderless Lakehouse so business users can query records across clouds in natural language; general availability of cross-cloud interconnect and caching for AWS and Azure, which Google Cloud said provides a private, secure connection with zero variable egress fees; and direct Iceberg access in AlloyDB, which lets users join live operational data with historical lakehouse data in a single PostgreSQL query without ETL pipelines or copies. Google Cloud said AlloyDB direct Iceberg access is up to 7x faster on aggregation queries. Harish Ramesh, staff data engineer at Delivery Hero talabat, said the company reduced data-movement costs by approximately 40% using the borderless Lakehouse.

[Google Cloud] Google Cloud announced the preview of AlloyDB Express, a new AlloyDB for PostgreSQL offering designed for developers and AI coding agents. According to Google Cloud, the service can create a ready-to-use PostgreSQL database in under two seconds via gcloud CLI, REST API, MCP, Terraform or the Console, and automatically pauses after 15 minutes of inactivity so no compute costs are incurred while idle, waking in seconds when a new query arrives. It includes built-in enterprise AI capabilities such as vector, text and hybrid search, along with security controls including IAM, VPC Service Controls and native PostgreSQL roles. Google Cloud said users can start with a free monthly allowance and pay only for what they use, and can upgrade an existing database to AlloyDB in one click while keeping data, schemas, users and connection strings intact, gaining enterprise high availability, read replicas and disaster recovery.

[Google Cloud] Mandiant responded in 2026 to multiple incidents in which threat actors hijacked enterprise access to large language models (LLMs), a tactic known as LLMJacking, according to Google Cloud. The intrusions spanned multiple cloud and LLM providers and affected organizations in North America, South America and Asia Pacific across financial services, retail, technology and transportation. Unlike the gradual resource consumption of cryptojacking, LLMJacking can generate large spikes in fraudulent cloud charges within hours; according to Google Cloud, fraudulent charges tied to resource consumption exceeded $100,000 in a single day in individual incidents, with cumulative charges at some organizations reaching $1 million before access was revoked. Google Threat Intelligence Group (GTIG) said it identified financially motivated threat clusters operating from multiple regions and is tracking a division of labor between actors establishing initial access and operators of downstream monetization infrastructure known as LLM Mixers, which it described as illicit routing gateways that proxy, aggregate and resell unauthorized LLM access. Google Cloud said the dominant initial access vector was exposed or stolen cloud identity tokens, including IAM credentials, service account keys, GitHub personal access tokens and active session tokens.

[Google Cloud] Google Cloud on November 2, 2026 published a monthly AI recap, saying it made several enterprise AI advances in October. According to Google Cloud, Gemini 4 Argon is now available on Google Cloud and was trained to maintain context across long, multi-step tasks, including complex software builds, financial and legal work, and cybersecurity tools such as CodeMender. Google Cloud said it announced general availability of an Anthropic-managed deployment option for Claude models, and said it is the only hyperscale cloud provider offering access to all three leading closed-source LLMs alongside more than 200 first-party, third-party and open models. Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity for Google Cloud customers, Gemini for government services is in preview, and Gemini Robotics Embodied Reasoning 2 is available on Google Cloud, according to the company.

[Google Cloud] The google-cloud-developer plugin is now available in the Build with Google section of the Antigravity Marketplace, according to Google Cloud, allowing users to install it directly starting November 2, 2026. Google Cloud said the plugin equips Antigravity with foundational skills, safety guardrails and live documentation grounding for building on Google Cloud. The bundled Developer Knowledge Model Context Protocol (MCP) server is pre-authorized by default, so users do not need to configure API keys, exchange OAuth tokens or edit configuration files, and the agent connects to official Google developer documentation immediately after installation, according to the company. Bundled skills include finding-google-skills, which discovers and loads skills on demand from the google/skills repository on GitHub; retrieving-developer-knowledge, which provides answers and code samples grounded in official Google documentation; gcloud, which validates syntax and applies safety checks to Google Cloud CLI commands; google-cloud-recipe-auth, which steers agents toward secure authentication patterns; and google-cloud-recipe-onboarding, which guides initial project setup and billing verification. Google Cloud said the agent requires user consent before enabling billable APIs or modifying cloud infrastructure.

[Google Cloud] Google Cloud published a technical blog post describing how to upgrade vector embedding models in AlloyDB AI with zero downtime. According to the post, upgrading embedding models requires re-embedding the entire dataset, rebuilding indexes and coordinating schema migrations without interrupting user queries; common pitfalls include incompatible vector spaces between models, index rebuild overhead, and DDL operations that acquire exclusive locks and can stall read and write transactions. Google Cloud said AlloyDB AI's model endpoint management provides in-database access to first-party, third-party or self-hosted embedding models, while auto embeddings generates new embeddings in the background at up to 3,000 per second and keeps them current as source data changes. ScaNN index builds are 10x faster than open source HNSW, according to the post. The walkthrough migrates the ecomm.product_reviews table from text-embedding-005 (768 dimensions) to gemini-embedding-001 (3,072 dimensions) while serving live traffic.

[Google Cloud] Google Threat Intelligence Group (GTIG) said that since late August 2026 it and Mandiant have observed multiple threat actors abusing the open-source, multi-agent pentesting frameworks ARTEX, PentAGI and CAIRN, as well as potentially custom harnesses, in incident response engagements, conducting actions across multiple phases of the attack lifecycle including lateral movement, database exploitation and exfiltration preparation. GTIG said its September 2026 report noted that some adversaries were beginning to move from basic prompting to agentic workflows and LLM-enabled automation. According to GTIG, such frameworks can autonomously explore multiple attack paths, troubleshoot errors and achieve objectives in seconds and minutes, increasing the speed and scale of intrusions. GTIG said the mechanics of the intrusions remain the same, relying on compromised credentials and unpatched systems, but automated chaining of complex attack paths introduces a new defensive challenge. Mandiant Consulting recommended control-plane isolation, short-lived credentials, least privilege, segmentation and behavioral detection as long-standing best practices.

[Google Cloud] Results from the new MLPerf Inference v0.5 benchmarks, which cover image classification, object detection and language translation, are now available, according to Google Cloud. Google Cloud said it submitted results for all three datacenter-class benchmarks using Cloud TPU v3 devices and demonstrated near-linear scalability up to a record 1 million images processed per second on ResNet-50 v1.5 using 32 Cloud TPU v3 devices. According to Google Cloud, Cloud TPUs offer bfloat16 floating-point numerics and let training and serving run on the same hardware platform, avoiding quantization, recalibration or retraining and the associated accuracy loss and engineering effort; Google Translate can push new models to production within hours of model validation, it said. Google Cloud said Cloud TPU v3 devices are publicly available to Google Cloud customers in beta and are also used across large-scale Google products including Google Search.

[Google Cloud] Google Cloud announced on November 13 the launch of Network Intelligence Center, describing it as its network monitoring, verification and optimization platform across cloud and on-prem data centers, alongside an initial set of modules. The company said growing adoption of multi-cloud and hybrid deployments is making networks more complex, leaving operations teams with limited visibility, long troubleshooting times and configuration errors discovered too late in production. The first phase includes four modules: Connectivity Tests and Network Topology, both in beta; and Performance Dashboard and Firewall Metrics & Insights in alpha, with more modules to follow. Google Cloud said Network Topology provides visibility into a customer's global GCP deployment and its interaction with the public internet, and lets users look back up to six weeks to track topology changes for diagnosis.

[Google Cloud] Google Cloud said on Nov. 19 it launched the Cloud Acceleration Program for SAP customers, an evolution of its earlier Lighthouse program, to help customers move SAP workloads from on-premises systems to Google Cloud or extend SAP solutions with Google Cloud and third-party technologies. According to Google Cloud, participants will gain new architecture templates, accelerators and SAP-focused support, plus partner-led assessment services, prototyping and SAP centers of excellence on Google Cloud. Accenture, Atos, Deloitte and HCL have created Google Cloud-specific business units, and along with Capgemini, DXC Technologies, Hitachi oXya, IBM Global Business Services, Infosys, NTT, TCS and Wipro will create SAP centers of excellence for Google Cloud. Atos said its Quick Cloud Start program can move a customer's SAP environment to Google Cloud in less than four weeks; Wipro developed a migration framework called Safe Passage to GCP; and SNP's CrystalBridge platform supports near-zero-downtime migrations.

[Google Cloud] Google Cloud, in its Oct 26-30 update roundup, said the newly released AlloyDB Omni web UI provides a centralized graphical dashboard that simplifies management of self-managed PostgreSQL workloads across hybrid, multi-cloud and air-gapped environments. The browser-based console gives database administrators real-time insight into cluster health, hardware utilization and active database connections, and includes built-in query diagnostics and performance tracking for AlloyDB's columnar engine, enabling troubleshooting without manual terminal commands, according to Google Cloud. Google Cloud also said native streaming support is now in public preview for API Gateway, allowing real-time and AI workloads behind existing gateways by routing Server-Sent Events, WebSockets, gRPC and chunked transfer encoding without modifying API specs, and strengthening integration with the Gemini platform. Google Cloud said it plans a Nov 12 Community TechTalk on architecting semantic model routing in Apigee.

[Google Cloud] Google Cloud said on October 30, 2026 that spend caps on Google Cloud Budgets are generally available, offering a way to protect workloads from runaway costs without risking resource deletion. According to Google Cloud, the release adds three updates inspired by feedback since the preview: a spend caps API accessible from the existing Cloud Billing Budgets API, allowing programmatic enforcement of financial boundaries for eligible services within each project; live cost reports that show the estimated costs that triggered a cap for AI services within minutes of usage; and automated monthly resets that restore blocked services at the start of the next month, while a manual override remains for governance policies requiring audit. Google Cloud recommends starting with strict, low-dollar caps on non-production environments such as developer sandboxes and testing clusters, and notes that default alerts go to billing administrators and project owners at 50%, 80% and 100% of the cap. The company said spend caps halt pay-as-you-go and CUD-covered requests once the threshold is hit, but Provisioned Throughput is never halted.

[Google Cloud] The Google Cloud AI Infrastructure team said it rolled out new products and updates in October to support AI and agentic workloads. Agent Compute Engine, a new compute service, provides hardware-isolated microVM sandboxes alongside governance capabilities including Agent Gateway and Agent Registry; according to Google Cloud, it offers sub-second activation, workload density through copy-on-write memory sharing, enterprise security and identity management, and cross-cloud observability. At NVIDIA GTC Berlin 2026, the team announced updates to its agentic infrastructure, including A5X instances powered by the NVIDIA Vera Rubin NVL72 platform, the Virgo Network scale-out fabric and Managed Lustre storage. For GKE, a CPU startup boost feature in preview uses Kubernetes In-place Pod Resize to raise container CPU resources during initialization and scale them back without restarts; Google Cloud said it accelerates cold starts by up to 2x. A preview of the Wiz Runtime Sensor add-on for GKE was also announced. Cloud Run added Temporal Serverless Workers that scale worker pools by task queue demand and down to zero when idle. In open source, llm-d 1.0 is a Kubernetes-native serving stack that unifies distributed LLM inference across TPU and GPU fleets, and Google, working with Meta, open-sourced TorchTPU to integrate TPUs into the PyTorch ecosystem.

[Google Cloud] Google Skills on October 30, 2026 released industry-specific video learning bundles for financial services and the legal profession, saying all videos are available at no cost. The financial services bundle comprises four courses covering financial advisor insights, portfolio resilience, bond issuance and credit market opportunities. The legal bundle comprises five courses on regulatory horizon scanning, data and DSAR responses, contract review and playbook creation, document redaction and NDA drafting. According to Google Cloud, legal teams can use Gemini to build compliance standards and review agreements against them, with the model flagging violations, explaining why they breach the standards and suggesting remedies. Google Cloud also said its no-cost Gemini Agent Ready (GEAR) program gives registrants 35 credits for interactive training environments on Google Skills.

[Google Cloud] Google Cloud detailed data cloud updates for October 19-23. According to Google Cloud, GoogleSQL in BigQuery now supports column aliases in SELECT and WHERE clauses (GA opt-in), reducing nested subqueries and simplifying migrations from Snowflake and Teradata, while cross-region table copying is now generally available to streamline regional data replication and disaster recovery. Conversational analytics in BigQuery expanded to Montréal and Toronto and now supports exporting agent responses to CSV or PDF. AlloyDB Omni 18.4.0 is GA with PostgreSQL 18.4, a query insights extension and Thales CipherTrust Manager support for transparent data encryption. Cloud SQL for PostgreSQL added managed connection pooling and now defaults to Private Service Connect for new instances in new projects. Google Cloud said digital identity platform Unico, after moving to BigQuery and Spanner, has prevented over $5 billion in fraud.

[Google Cloud] Google Cloud said its Knowledge Catalog now generally supports expanded semantic model ingestion, natively pulling semantic models and metric definitions from Looker LookML, BigQuery Graph (including BigQuery measures), dbt MetricFlow, Snowflake Semantic Views, Databricks Metric Views and SAP Core Schema Notation (CSN). The definitions are connected with schema, lineage, data profiles, business glossary and other metadata in a single context graph without moving data or rewriting business logic in a new language, according to Google Cloud. The company said internal benchmarks show grounding agents in Knowledge Catalog makes them nearly 4x more accurate, with up to 33% lower latency and up to 75% fewer tokens. When two tools define the same metric differently, both definitions appear with their provenance so data stewards can reconcile the conflict before it reaches end users, Google Cloud said.

[Google Cloud] Google Cloud has added a $250,000 Root-in-Cloud reward tier to its Vulnerability Reward Program (VRP) to recognize high-impact architectural security research, according to Chris Betz, CISO of Google Cloud. According to Google Cloud, the tier requires researchers to demonstrate full administrative control, equivalent to roles/owner or complete IAM control, over arbitrary Google Cloud organizations or projects with no prior access or special project setup. The exploit path cannot depend on pre-existing access or footholds, leaked, phished or stolen credentials, or user interaction or social engineering, Google Cloud said. Valid reports showing remote code execution in Google's underlying production infrastructure (Borg) or achieving root-in-prod will automatically be evaluated by the Google Cloud VRP panel for root-in-cloud eligibility, according to the company. Researchers must halt all probing immediately upon achieving any execution in Borg; escalation, extracting production credentials and lateral movement are not allowed and will result in disqualification, Google Cloud said.

[Google Cloud] Google Cloud said its borderless Lakehouse now lets BigQuery query Databricks-managed Apache Iceberg tables directly, without copying data. According to Google Cloud, BigQuery reads the same files Databricks writes, and new changes appear after Lakehouse's next scheduled metadata refresh, which can run as often as every five minutes. The approach extends federated queries to external clouds such as Amazon S3 and Azure Blob Storage through the open Apache Iceberg REST Catalog and cross-cloud connections; remote-cloud egress charges still apply, but Cross-Cloud Interconnect (AWS) and Lakehouse caching can reduce them, Google Cloud said. On governance, administrators can configure a BigQuery Cloud Resource Connection tied to a service account that manages IAM rights for GCS buckets, enabling row-level and column-level security and dynamic data masking on external Iceberg tables. Google Cloud also said Databricks-managed Iceberg tables do not support partition transforms and that Databricks recommends liquid clustering instead.

[Google Cloud] Google Cloud announced the launch of XREAL AURA, which it described as a first-of-its-kind set of wired XR glasses built on Android XR. According to Google Cloud, the device uses a newly introduced split-compute architecture for spatial computing on wired glasses, features an optical-see-through display with a 70-degree field of view that layers visual content onto the physical world, and has a glasses frame weight of less than 95g for extended work sessions. AURA offers out-of-the-box support for enterprise apps available on the Google Play Store, and its DisplayPort-in function lets users connect laptops across platforms alongside Android apps. Google Cloud said XREAL AURA devices will be managed through the Android Enterprise framework, with features including Managed Google Play, Android zero-touch enrollment, fully managed device mode, Google Play Protect, core Android platform security and Device Trust. The device is available only for users in the United States, Canada, Japan and South Korea.

[Google Cloud] Google Cloud said its Google Agent Skills repository has passed 21,000 GitHub stars, up from 15,000, and now hosts more than 170 skills from teams across Google, distributed across multiple AI platforms. According to Google Cloud, the team ships every skill update like a product release, requiring strict semantic versioning, deterministic automated validation and mandatory evaluation evidence. Google Cloud said a CI/CD pipeline triggers when an author submits a change, and a skill must pass six automated gates plus human approval before export to GitHub: structural conformance, skill linters, semantic versioning, evaluation suite, link validation and an AI review agent. Google Cloud said a skill steers an AI agent's reasoning loop, tool selection and execution path, so a misplaced phrase can cause invalid command flags, an infinite loop or wasted tokens, making every instruction edit a behavioral change.

[Google Cloud] Google Cloud said it is previewing fully managed blue-green deployments in Cloud SQL, starting with Cloud SQL for MySQL. According to Google Cloud, users can run major version upgrades and configuration updates with minimal downtime through the console or automated CLI scripts, while a parallel environment is continuously synchronized so teams can test with production data before cutover and keep the original database as a fallback. Google Cloud said organizations using Cloud SQL Enterprise Plus can achieve sub-second downtime for maintenance, and that the switchover automatically transfers the existing database IP address, instance name and backups without application changes.

[Google Cloud] Google Cloud said its public sector team, working with NASA's Human Research Program, Johnson Space Center Robotics Team and the Jet Propulsion Laboratory, is developing a proof-of-concept physical robotic system aimed at supporting deep space missions and assisting astronauts. The system is designed as a versatile payload handler for cargo, logistics and equipment, and can pivot to medical support, such as retrieving and preparing an ultrasound machine for a crew member with suspected kidney stones. In a demonstration at Johnson Space Center's iMETRO facility, the system performed a pick-and-place maneuver to locate and move a medical bag; using an eye-to-hand camera configuration, the open-source Pi0 model developed on Google Cloud infrastructure achieved an initial 85% baseline success rate in early lab environments, according to Google Cloud.

[Google Cloud] Gemini Robotics Embodied Reasoning 2 (ER 2) is now available in public preview on the Gemini platform, according to Google Cloud, letting organizations pair Google DeepMind's embodied reasoning model for robotics with Google Cloud's enterprise-grade security, multi-agent orchestration and end-to-end lifecycle management to simulate, evaluate and deploy physical AI at scale. The model works across robot embodiments from stationary arms and wheeled rovers to quadrupeds and humanoids, supports multi-robot collaboration and natural-language interaction, and lets developers declare low-level control interfaces such as vision-language-action models and navigation APIs as tools. According to Google Cloud, it pinpoints the exact video frame where a critical event occurs with 91.3% accuracy and a 0.96-second mean absolute frame distance, running at 4x the execution speed of many larger model categories.

[Google Cloud] Google Cloud announced the general availability of the AlloyDB Omni web UI for Red Hat and containers. According to Google Cloud, the production-ready, read-only console gives administrators direct visibility into hardware telemetry, query performance and configuration flags, so they can troubleshoot workloads without relying on command-line tools. AlloyDB Omni is a downloadable, self-managed edition of AlloyDB that runs across on-premises data centers, edge environments and multi-cloud deployments, with built-in columnar acceleration and integrated AI tooling. The web UI offers cluster overview and connection-profile switching, real-time CPU, memory and disk I/O monitoring, query analysis that ranks execution-heavy SQL statements by latency and call frequency, Columnar Engine memory and cache-coverage diagnostics, and runtime parameter auditing to spot configuration drift. Google Cloud said the interface operates strictly in read-only mode, so diagnostic sessions cannot execute write commands or modify underlying database configurations.

[Google Cloud] Google Cloud and Oracle said they are partnering to bring Google Tensor Processing Units (TPUs) to Oracle Cloud Infrastructure (OCI), which according to Google Cloud will become the first hyperscaler outside Google to offer TPUs. The two companies will deploy TPU capacity on OCI starting in calendar year 2027. According to Google Cloud, the latest-generation TPUs deliver up to 80% better performance per dollar for low-latency serving and sustain over 97% goodput across thousands of chips on large-scale training runs. Google Cloud also said TorchTPU, its native PyTorch stack for Google TPUs, is now available as open source on GitHub.

[Google Cloud] Google Cloud said in a blog post that as AI workflows shift from static requests to autonomous agents, demand for secure, high-density sandboxes is pushing conventional infrastructure to its architectural limits, with some research labs requiring up to one million concurrent code-execution sandboxes. According to Google Cloud, GKE Dataplane V2, powered by eBPF, maps pod labels to a numeric identity to enforce network security policies. To address slow pod creation caused by serial identity allocation for highly unique per-pod labels, the company said it separates identifying data from functional metadata, moving high-cardinality fields from Pod Labels to Annotations and calculating network identities from labels rather than annotations, bypassing per-pod identity generation. Google Cloud said Cilium label exclusion filters can further control identity counts. According to Google Cloud, a customer said it used Agent Substrate on GKE to orchestrate hundreds of thousands of secure environments across clusters and handle spikes of more than 30,000 sandboxes on a single cluster.

[Google Cloud] BigQuery measures is now generally available, unifying business intelligence metric definitions with knowledge graphs in a single model, Google Cloud said. According to Google Cloud, organizations needing both BI and knowledge graphs have historically run two siloed stacks, bridged by brittle data pipelines and runtime reconciliation by analysts and AI agents, which the company said is where data governance broke down. Built on BigQuery Graph, BigQuery measures lets users define measures directly on the graph entities they describe, so dashboards and AI agents get a trusted number along with the relational context behind it, Google Cloud said. Amanda Reelick, Head of Data Enablement at Woolworths NZ, said the capability ensures Looker dashboards, conversational agents and data team queries operate on identical logic, guaranteeing metric alignment and reducing data team maintenance overhead.

[Google Cloud] Google Cloud published a technical guide on securely connecting Gemini Enterprise to Snowflake using the Model Context Protocol (MCP) and the Google Agent Development Kit (ADK). According to Google Cloud, the architecture lets business users ask questions in natural language; the ADK agent passes the question to a Snowflake Cortex Agent exposed through the Snowflake MCP server, which translates it into SQL via the Cortex Analyst semantic view and executes it inside Snowflake's governed boundary using the authenticated user's OAuth session and role privileges, returning structured results to the agent. Google Cloud said the approach provides deterministic text-to-SQL translation and reduces the risk of schema hallucinations or unvalidated SQL execution. The guide describes this custom orchestration pattern as aimed at developers building complex multi-agent systems on Google Cloud, while standard enterprise use cases can use joint solutions.

[Google Cloud] Google Cloud said it is expanding its partnership with Oracle across the AI stack, spanning AI compute and infrastructure, including making Google TPUs available through Oracle Cloud Infrastructure, enterprise data and agentic AI, and deepening integration between Oracle AI Database@Google Cloud, Google's Agentic Data Cloud and Gemini Enterprise. The companies announced the move at Oracle AI World in Las Vegas. According to Google Cloud, the Remote Model Context Protocol Server for Oracle workloads, powered by the IAM-governed ExecuteSQL Data API, is now generally available, giving AI agents and Google Cloud services a standardized interface to securely run SQL and interact with Oracle databases. The Knowledge Catalog integration is in preview, unifying metadata discovery and governance across Oracle databases and the broader Agentic Data Cloud. Google Cloud also announced a preview of the Gemini connector for Oracle AI Database@Google Cloud, letting users query Oracle databases in natural language inside the Gemini app, and said Oracle in Gemini Enterprise is available today with conversational analytics, cross-cloud database connectors and domain-aware agents, anchored by more than 135 pre-built operational and developer skills.

[Google Cloud] Google Cloud said on October 27 that its VMware solution on Google Cloud is now generally available as part of the Google Cloud Modernize portfolio. The self-managed offering runs VMware Cloud Foundation (VCF) on dedicated Google Cloud bare metal and gives customers full root administrative access across ESXi, vCenter, NSX Manager and other VCF components, with control over maintenance windows and custom integration with third-party monitoring and management tools. According to Google Cloud, Broadcom's launch of VCF 9.x is a key driver behind the self-managed option, whose unified VCF control plane supports end-to-end stack deployment and lifecycle management. Abhay Kumar, global head of hyperscalers in Broadcom's VMware Cloud Foundation Division, said VCF license portability lets customers bring existing subscriptions and run VMware workloads unchanged on dedicated Z3 bare metal, with Google Cloud commitments and discounts available to optimize costs. The offering sits alongside the Google-managed VMware Engine, which runs VCF 5.2 with automated patching, updates and monitoring handled by Google.

[Google Cloud] Google Cloud said on October 26 that LookML semantic language indexing in Knowledge Catalog for Looker Core on Google Cloud is now generally available. According to Google Cloud, as enterprises move AI agents from prototypes into production, data teams are finding that access to data is not the same as understanding it; answering complex business questions accurately requires context such as table schemas, data lineage, governance policies and business definitions. Google Cloud said Knowledge Catalog provides a unified context graph aggregating metadata across structured tables and unstructured documents, while LookML encodes production-tested semantic logic including multi-grain join protections, conditional filters and exact KPI formulas. With the native integration, the LookML business logic behind trusted BI reporting is automatically indexed into Knowledge Catalog, giving Gemini Enterprise and custom AI agents both enterprise-wide discovery and deterministic BI semantics, according to Google Cloud. The company said that without explicit semantic rules, agents can fall into metric definition drift and naive join fan-outs that inflate financial totals.

[Google Cloud] Google Cloud on October 26 introduced IAM Helper, a suite of intelligent, agent-facing building blocks that it said can automate complex, end-to-end IAM workflows. According to Google Cloud, the product acts as an agent-native orchestration and interface layer for Google Cloud IAM, augmenting core IAM engines including Policy Intelligence and Privileged Access Manager (PAM) so that IAM capabilities can be called directly by autonomous code assistants, deployment pipelines and custom agents. Its Agent Skills are packaged using the open Agent Skills standard and work across common agents including Gemini CLI, Gemini Cloud Assist, ADK, Claude Code and Cursor, Google Cloud said. The company said built-in guardrails separate read-only discovery from policy changes: diagnostics and simulations run automatically, while changes to access policies pause for human-in-the-loop confirmation based on customizable risk tiers.