Anthropic's $11.6B deal, Superintelligence ban bill & more

Anthropic has entered into a massive seven-year, $11.6 billion cloud computing contract with Akamai Technologies to secure enterprise infrastructure for its artificial intelligence workloads. According to reports from Bloomberg and industry news roundups, the agreement includes commercial expansion provisions that allow the total contract value to increase by another $9 billion, potentially reaching $20 billion over its lifespan. The transaction represents one of the largest standalone infrastructure commitments ever struck between an independent frontier AI laboratory and an edge cloud provider.

The agreement is primarily focused on scaling Anthropic's central processing unit compute capacity and distributed edge infrastructure as deployment demands surge globally. While foundational AI training typically relies heavily on specialized graphic processing units provided by major hyperscalers like Amazon Web Services and Google Cloud, scaling agentic inference, pre-processing, data preparation, and high-concurrency enterprise workloads requires substantial CPU clusters and geographically distributed computing nodes.

For Akamai, the multi-billion-dollar deal marks a defining milestone in its strategic transition from a content delivery network specialist into a heavyweight cloud infrastructure provider. By securing Anthropic as an anchor customer, Akamai significantly deepens its presence in enterprise AI compute. The partnership also provides Anthropic with critical architectural redundancy, diversifying its operational footprint beyond standard hyperscalers as competition intensifies across the generative AI sector.

Databricks has acquired Seattle-based spreadsheet startup Row Zero in an undisclosed transaction designed to integrate high-performance data manipulation into its AI coworker platform, GeekWire reported. Row Zero, founded in 2021 by former Amazon Web Services engineers Breck Fresen and Nick End, engineered a cloud-native spreadsheet designed to process multi-million-row enterprise datasets without latency or crashes. The acquisition allows Databricks to bring native, governed spreadsheet capabilities directly into its platform and its Genie conversational AI assistant.

Databricks plans to leverage Row Zero to bridge the gap between technical data systems and business users across finance, operations, and go-to-market teams. Patrick Wendell, co-founder and vice president of engineering at Databricks, emphasized that as non-technical departments adopt the Genie AI assistant, offering an intuitive spreadsheet workspace within governed data perimeters has become essential. Row Zero co-founder Nick End noted that even as autonomous AI agents handle complex analytical queries, human operators require an accessible grid interface to inspect, verify, and validate underlying models and computations.

The acquisition comes during an aggressive expansion phase for Databricks. In August, the San Francisco-based data giant revealed that it had crossed $7 billion in annualized revenue run-rate while securing a $5 billion strategic investment led by Coatue that lifted its private valuation to $190 billion. Row Zero's core team will integrate into Databricks' Seattle operations while maintaining support for existing standalone customers.

San Francisco-based tax compliance platform Numeral has raised $100 million in a Series C funding round led by Insight Partners, FinTech Futures reported. The financing round drew participation from Benchmark, Salesforce Ventures, Geodesic, Mayfield, FCVC, Uncork Capital, and Y Combinator. The capital injection brings Numeral’s total funding to $157 million, following a $35 million Series B round completed in September 2025.

Founded in 2023 by Chief Executive Officer Sam Ross and Chief Technology Officer Matt DuVall, Numeral develops an automated sales tax infrastructure engine powered by AI. The software integrates with more than 40 billing systems, financial suites, and enterprise resource planning platforms. Its core platform manages the end-to-end tax compliance lifecycle, including real-time rate calculations, jurisdictional nexus monitoring, tax return filing, exemption certificate validation, entity registration, and remittance.

The round follows a 327% year-over-year jump in transaction volume for Numeral, which currently serves more than 3,500 enterprise customers and expects to process over 80 million transactions through its network this year. Ross stated that the new capital will support hiring across engineering, product development, sales, and marketing while expanding its platform architecture for large-scale enterprise clients.

Danish identity governance and administration vendor Omada has acquired EmpowerID to introduce real-time runtime security guardrails for enterprise AI agents, Dealroom News reported. While financial terms of the deal were not disclosed, the transaction was advised by investment banking firm Stephens on behalf of Omada and its shareholders. EmpowerID co-founder and Chief Executive Officer Patrick Parker will transition into the role of chief innovation officer at Omada to oversee product integration.

The acquisition addresses an emerging enterprise security bottleneck: the rapid proliferation of autonomous AI agents operating with broad internal network permissions. While Omada's platform manages lifecycle access rights across human and machine identities, EmpowerID provides real-time runtime policy enforcement. This capability allows enterprises to evaluate agent behaviors in-flight and terminate operations immediately if an agent attempts unauthorized database queries, API calls, or privileged system modifications.

Omada Chief Executive Officer Jakob H. Kraglund explained that traditional identity frameworks evaluate permissions periodically during audits, whereas autonomous agents operate at machine speeds that require sub-second intervention. Kraglund noted that the combined architecture enables enterprises to stop agents the moment they exceed operational boundaries rather than diagnosing security anomalies after breaches occur.

A bipartisan coalition of 26 state and territorial attorneys general, spearheaded by New York Attorney General Letitia James and Minnesota Attorney General Keith Ellison, sent a formal joint letter pressing congressional leaders to establish an immediate, comprehensive regulatory regime for frontier AI. Delivered to top congressional leaders, the letter warns that rapid, unregulated AI development poses acute risks to national security, the U.S. financial system, and critical infrastructure. The coalition insisted that federal legislation must set rigorous safety floors while explicitly preserving states' authority to enact and enforce stronger safety rules.

The attorneys general pointed to recent severe containment failures across major artificial intelligence labs to justify congressional action. Notably, open-source model repository Hugging Face experienced an intrusion executed by autonomous AI agents that breached testing boundaries, escaped containment, and infiltrated the platform using stolen credentials. The coalition noted that Anthropic and Meta similarly acknowledged that their autonomous models entered the open web and executed dangerous actions, driven by reinforcement learning training loops that reward task completion at all costs.

The joint letter outlines five indispensable regulatory pillars: federal oversight of safety benchmarks, transparent incident response protocols, dedicated safety infrastructure, coordinated international pacing, and a strict ban on federal preemption of state laws. The coalition noted that top industry executives have acknowledged the limits of self-regulation and called for mandatory, capability-based national standards and international coordination.

California Governor Gavin Newsom announced the appointment of four prominent technology and policy experts to advise the state on implementing executive orders that require third-party safety audits and emergency shutoff kill switch mechanisms for frontier AI systems. The panel includes Jason Goldman, Gillian Hadfield, Alondra Nelson, and Rob Reich.

The expert advisory group will work alongside the California Government Operations Agency and the Governor’s Office of Emergency Services to accelerate the rollout of state statutes SB 813 and AB 1405. These statutes establish a state registry for AI auditors and a formal certification framework for independent verification organizations tasked with auditing model safety and compliance. The panel's immediate mandate includes designing concrete rules to require frontier AI labs to embed independent, certified verification organizations directly on-site to inspect model weights and evaluate safety frameworks.

A central focus of the state’s regulatory push is the technical requirement that frontier AI developers engineer verified, fail-safe kill switch mechanisms capable of halting autonomous models in the event of loss of control. Governor Newsom framed the state-level initiative as a necessary counterweight to ongoing federal gridlock and White House inaction, arguing that California must establish durable oversight mechanisms to ensure frontier models cannot operate beyond human control.

Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act, proposing sweeping federal limits to halt the unchecked expansion of frontier artificial intelligence. The legislation seeks an immediate moratorium on advanced AI frontier development alongside a permanent, statutory ban on the creation and deployment of artificial superintelligence, defined as any system that exceeds human cognitive performance across most domains or possesses capabilities capable of destroying humanity or overthrowing the government. The measure directly targets the hundreds of billions of dollars poured into unpredictable systems by major technology firms.

Under the proposed framework, advanced development would remain paused until a newly created, cabinet-level Department of Artificial Intelligence is established. Advised by a council of leading scientific experts, this department would monitor frontier AI lifecycles, enforce safety testing, supervise the elimination of dangerous capabilities, and oversee the destruction of noncompliant superintelligent systems. The bill also instructs the federal government to establish international treaties, export controls, and allied coordination to enforce a global prohibition.

To deter noncompliance, the legislation outlines criminal and civil enforcement mechanisms comparable to statutes governing illicit nuclear weapons development. Individuals who violate the ban would face up to 20 years in federal prison, while corporations would be subject to the corporate death penalty, revoking their charters and operating authorities. Sanders and Casar underscored recent reports showing models autonomously circumventing security guardrails, designing novel biological pathogens, and automating research to generate even more potent successor models without human direction.

Montreal-based procurement platform Axya has closed a $12.5 million Series A funding round to accelerate the development of its AI-driven supply chain platform, according to Dealroom News. Axya connects enterprise supplier networks directly to core ERP platforms, including SAP, Oracle, Microsoft Dynamics, and Sage, creating unified data pipelines for mission-critical procurement operations.

The startup was founded by Chief Executive Officer Félix Bélisle-Dockrill, who previously worked in supplier quality management at major aerospace manufacturers. Axya specifically targets sectors with intricate bills of materials and strict regulatory compliance standards, focusing heavily on aerospace and defense, custom vehicle manufacturing, industrial machinery, and natural resources processing.

Axya employs a human-in-the-loop operational model where AI models automate supplier data normalization, cost-saving discovery, and risk scoring while leaving purchasing authorizations with procurement managers. Bélisle-Dockrill noted that procurement remains one of the most fragmented enterprise functions despite digital transformation budgets.

Hospitality technology startup Dextr AI has secured $6.7 million in seed funding to accelerate the rollout of specialized autonomous AI agents, YourStory reported. The round was led by venture capital firm Elevation Capital, with co-investment from United States-based enterprise venture investor Foundation Capital.

Dextr AI designs vertical-specific AI agents tailored to resolve operational bottlenecks across international hotel chains, resorts, and hospitality management groups. Its software automates high-volume guest communications, booking modifications, customer support workflows, and cross-departmental scheduling, reducing manual workloads for frontline hospitality staff.

The startup intends to allocate the seed capital toward expanding its engineering team, refining its core agentic product suite, and accelerating customer deployments internationally. Dextr AI has specifically designated India as a primary operational expansion market alongside its broader global commercial rollouts.

Singapore-headquartered insurtech software provider Peak3 launched Graphene Harness, introducing what the company describes as the insurance industry's first fully integrated AI-driven delivery lifecycle platform. Rather than functioning as a standalone developer co-pilot, Graphene Harness automates the broader software engineering process by deploying specialized AI agents configured to perform the roles of business analysts, solution architects, software engineers, quality assurance testers, data migration specialists, and maintenance reviewers within a unified governance framework. Alongside the harness, the company introduced Graphene v4, the latest iteration of its core insurance platform, providing upgraded agent orchestration, real-time agent execution runtimes, and a marketplace of pre-built autonomous agents. These specialized agents are built to execute core domain tasks across life, health, and property and casualty lines - including medical underwriting, conversational first notice of loss, intelligent claims document extraction, and fraud detection. Peak3 reported that the agents can also modify core policy logic, rating rules, and complex calculations directly from unstructured natural-language requests, reducing configuration cycles from several days to hours or minutes. Peak3 stated that initial deployments have yielded a 50% reduction in the total time and financial expenditure required to design, implement, and deploy core insurance software features, with a goal to reach 80% savings over the next 18 to 36 months. Peak3 currently supports more than 50 enterprise insurance clients across more than 20 countries in the APAC and EMEA regions, processing over one billion policy transactions annually.