U.S. Pulls Plug on Anthropic Models as Global Tech Rules Tighten

AI & Machine Learning

Bezos‑backed Prometheus closed a headline‑grabbing $12 billion round at an estimated $41 billion valuation, a capital infusion the company says will accelerate its frontier model development and enterprise LLM offerings. The funding, reported by GeekWire, underscores continued deep-pocketed investor appetite for companies building foundational models and competing in the enterprise AI market, and it signals a push to scale compute, data partnerships, and commercial productisation. For competitors and cloud providers the raise intensifies pressure to secure hosting deals and differentiated tooling to win large enterprise customers. The deal also raises questions about governance and safety oversight as better‑funded labs push faster on model capabilities and deployment. Source: GeekWire Verified: True

A new arXiv preprint dubbed “Human‑on‑the‑Bridge,” covered by Let’s Data Science, proposes a scalable, human-in-the-loop evaluation paradigm for agentic AI that embeds oversight into long‑horizon interactive testing. The paper argues existing benchmarks fail to capture emergent, interactive failure modes and suggests reproducible protocols where human supervisors intervene at critical junctures to measure system behavior under realistic constraints. If adopted, the framework could become a practical audit tool for both labs and regulators wanting to stress-test agentic systems before wide deployment. The submission arrives amid heightened scrutiny of agentic capabilities and complements other efforts to make safety evaluations more systematic and externally reproducible. Source: Let’s Data Science Verified: True

Consumer Hardware

Leaks collected by 9to5Google suggest Valve is preparing a late‑June reveal of a renewed “Steam Machine” console, showing supply‑chain images and tentative launch timings that point to a hardware push back into living rooms. The reporting highlights components, design hints and the possibility of Valve partnering with third‑party OEMs for different SKUs, indicating the company may target both PC gamers and a broader console audience. If the timeline holds, retailers and accessory makers will have only weeks to prepare and Valve will need to articulate a clear software and content strategy to avoid fragmenting its existing Steam Deck and PC ecosystem. The leak-driven nature of the story means some details may shift, but the market reaction will be telling for hybrid PC-console initiatives. Source: 9to5Google Verified: True

Mashable’s roundup on the Galaxy Z Fold 8 collects design leaks and supply‑chain whispers suggesting Samsung plans iterative hinge and display refinements for a summer launch, positioning the Fold 8 as a marquee device for premium handset cycles. The coverage synthesises rumored spec bumps, improved durability claims and potential camera module changes that aim to make the foldable more mainstream. For Samsung, hitting reliable hinge performance and battery life targets will be crucial to expand the foldable category beyond early adopters. Analysts will watch pricing and carrier incentives closely, since those will determine whether the Fold 8 can broaden foldable adoption this year. Source: Mashable Verified: True

Cybersecurity

CISA issued Binding Operational Directive BOD‑26‑04 directing federal civilian agencies to prioritise and accelerate high‑risk security updates according to a risk‑based framework, a move intended to close longstanding patching gaps across government. The directive specifies timelines and reporting expectations that aim to force faster remediation of critical vulnerabilities, reflecting growing concern about exploitation windows and nation‑state targeting of federal assets. Operationally, agencies will need tighter inventory, telemetry and patch orchestration to comply, likely driving procurement and staffing changes in the near term. The policy could also cascade into state and private sector expectations as contractors align to federal risk‑based update norms. Source: CISA Verified: True

Enterprise Infrastructure

FedScoop reports AWS won the DHS “Cumulus” cloud contract, an estimated $2.6 billion, marking a major consolidation of cloud services across Department of Homeland Security components and cementing AWS’s position in federal cloud work. The multi‑year deal covers a broad set of services and is described as a strategic consolidation that will affect how DHS standardises tooling, identity, and secure access across disparate offices. For competitors, the award highlights the continuing importance of federal procurement scale and security certifications; for DHS it raises questions about vendor lock‑in and migration plans. The contract will likely accelerate modernization projects inside DHS but will also draw attention from oversight bodies focused on resilience and multi‑cloud risk mitigation. Source: FedScoop Verified: True

Parallel Works announced an expansion of its ACTIVATE control plane to give defense and research communities unified orchestration across hybrid HPC and cloud resources, promising simplified workflows for large training and simulation jobs. The HPCwire report describes features that abstract scheduling, data movement and cost controls across on‑prem clusters and multiple cloud providers, which could materially reduce friction for defense researchers running mixed workloads. If adopted, the toolset could speed experimental cycles and lower entry barriers for smaller research groups needing burstable cloud capacity. The offering also underscores a broader trend: software layers that mask heterogeneity and let users treat HPC and cloud as a single elastic resource. Source: HPCwire Verified: True

Investing.com reports ARK Venture Fund invested in Hydra Host, a startup building a marketplace for AI GPU capacity that aims to connect buyers of training/inference cycles with specialised providers. The funding signals investor interest in marketplace models that can flexibly allocate scarce accelerator resources and provide richer telemetry and SLAs than spot purchases through hyperscalers. For enterprise AI teams, such marketplaces could offer lower costs or specialised hardware (e.g., custom memory stacks) that standard cloud offerings do not provide. The move also highlights how GPU demand has spawned adjacent infrastructure businesses that seek to commoditise access while adding management layers and compliance features. Source: Investing.com Verified: True

Policy & Regulation

TechCrunch reports the U.S. government ordered external access to two of Anthropic’s newest models, Claude Fable 5 and Claude Mythos 5, to be cut following safety warnings from Anthropic itself and concerns about national‑security and misuse. This formal directive represents a significant escalation in how regulators and national security agencies can intervene in the deployment of frontier models, moving beyond voluntary lab guidance to enforced access restrictions. The step raises urgent questions about criteria, transparency, and the mechanisms governments will use to limit deployed AI capabilities while balancing innovation and oversight. It also signals to other labs that safety flags can trigger direct government action, potentially reshaping deployment strategies and compliance investments. Source: TechCrunch Verified: True

The U.K. government unveiled a proposal to ban children under 16 from using mainstream social networks, requiring platforms to implement robust age verification and change onboarding and recommendation algorithms to protect young users. TechCrunch’s coverage frames the move as part of a broader child‑safety regulatory package that would force major platforms to redesign key features and face enforcement penalties for non‑compliance. If enacted, the policy would create significant engineering and privacy design challenges for platforms that rely on lightweight signups and algorithmic feeds, while likely prompting legal and technical debates over age verification methods. The proposal also signals growing international momentum toward stricter platform regulation focused on minors, and platforms will need to weigh compliance costs against user engagement impacts. Source: TechCrunch Verified: True

India ordered a temporary nationwide restriction on Telegram and asked the app to disable message‑editing features while investigators probe coordinated cheating on national exams, a block set to last until June 22 per TechCrunch’s reporting. The move illustrates how governments can leverage network‑level blocks and feature restrictions to pursue criminal or administrative investigations, but it also raises concerns about collateral impacts on free expression and emergency communications. For messaging platforms, the incident underscores the ongoing tension between strong privacy/feature sets and compliance with local law enforcement demands. The day‑to‑day operational realities of such temporary blocks will influence how global apps architect compliance and feature toggles for high‑risk jurisdictions. Source: TechCrunch Verified: True