Anthropic secures $10B compute; DeepSeek hikes prices

Anthropic / Claude ecosystem

Anthropic Improves Fable 5 Biology Safeguards, Cutting False Positives

Anthropic updated Claude Fable 5's biology safeguards, reducing fallbacks by approximately 85% while maintaining protection against dual-use content. The update increases handling of everyday biology queries without compromising safety.

Anthropic and Karnataka Government Discuss AI Partnership for Public Services

Anthropic's delegation met with Karnataka's Chief Minister to explore opening state datasets for AI-powered public solutions. Discussions included building an AI ecosystem with an AI university, skilling programs, and startup support.

Anthropic Signs $10 Billion Compute Deal with Volta Infra Holdings

Anthropic has signed a $10 billion deal over six years with Volta Infra Holdings, a stealth-backed Norwegian cloud startup. The deal centers on a 133 MW data center powered by Vera Rubin chips.

EU Lawmakers Examine Claude Mythos within AI Safety Rules

Anthropic representatives discussed AI safety with EU lawmakers, focusing on how Claude Mythos's advanced cybersecurity capabilities fit within existing EU AI Act and cybersecurity regulations.

Claude Code Introduces Self-Hosted Environments in Public Beta

Anthropic launched a public beta for running Claude Code sessions on customer-controlled infrastructure. This offers data locality and compliance benefits while maintaining integration with development workflows.

Claude Code RCE Vulnerability Allows Malicious Pull Requests to Execute Commands

A remote code execution flaw in Claude Code allows attackers to trigger commands via crafted pull requests during initialization, bypassing prompt approval and trusted workspace safeguards.

Anthropic Establishes In-House AI Chip Development Team

Anthropic has formed an internal team to design Claude-specific AI chips, aiming to improve efficiency and performance while maintaining a multi-chip strategy with AWS, Google, Nvidia, and AMD.

Frontier model providers

DeepSeek Plans Significant API Price Increase

DeepSeek announced a forthcoming price hike for its API services, with exact effective dates pending. The increase will affect billing structures for high concurrency and complex reasoning models.

Amazon Australia Launches Alexa+ with Advanced AI Capabilities

Alexa+ launched in Australia, bringing AI-powered features such as dinner reservation booking, gift ordering, price tracking, and home management through multi-app integration.

AI developer tooling & infrastructure

Claude Code and Gemini CLI Flaws Expose CI Workflow Secrets via GitHub Issues

Vulnerabilities (CVE-2026-12537, CVE-2026-54316) in Claude Code and Gemini CLI allow GitHub issues to reach CI workflow secrets, affecting multiple versions. Fixes have been released.

OpenAI and Rivals Agree on Agent Plugins Standard for Cross-Platform AI Agents

Five AI companies, including OpenAI, have agreed on a unified packaging standard called Agent Plugins, enabling AI agent extensions to run across multiple platforms.

Trust3 AI Launches Research Labs to Secure Entire Agent Lifecycle

Trust3 AI Labs was launched to address security across the agent lifecycle, covering five pillars from development to runtime for agentic enterprises.

Cloud & platform providers

AWS Announces Fractional GPU Scheduling on ECS with G6f Instances

Amazon ECS now supports fractional GPU scheduling on EC2 G6f instances, allowing right-sizing of GPU workloads and reducing costs.

AI policy, regulation & governance

Washington's AI Rulebook Kept Private, Frustrating Smaller Labs

The US government's frontier AI framework, which includes voluntary pre-release review, remains undisclosed to the public, prompting criticism from smaller AI labs about transparency and fairness.

Australian Signals Directorate Flags Foreign Control of AI Vendors as Board-Level Risk

The ASD issued guidance urging company boards to treat foreign control of AI vendors as a cyber risk, recommending stronger governance and security posture to counter frontier-model threats.

Industry & market moves

AMD Acquires Taalas to Boost AI Inference Capabilities

AMD announced the acquisition of Taalas, a startup specializing in hard-wiring AI models into silicon, to enhance its AI inference accelerator roadmap.

DeepSeek Invests in Unitree to Develop AI Brain for Humanoid Robots

DeepSeek is participating in Unitree's Shanghai IPO and will collaborate to develop AI models for humanoid robots, aiming to create an 'AI brain' for the robots.

Hotelmize Raises Series B from Alibaba Hong Kong Entrepreneurs Fund for AI-Driven Revenue Management

Hotelmize, an AI-powered hotel booking optimization platform, raised a Series B round led by Alibaba Hong Kong Entrepreneurs Fund and Brilliance Ventures to expand in APAC.

AI product & feature launches

Acceldata Adds AI Observability to xLake Platform

Acceldata introduced AI Observability to its xLake platform, enabling monitoring of prompts, model calls, tool use, and agent activity alongside data quality and lineage across on-prem and cloud.

BioNexus Showcases 'AI Scientist Agent' for Drug Discovery, Seeks Seed Funding

BioNexus demonstrated its AI Scientist Agent for drug discovery and is pursuing seed investment. The company has partnered with 15 key institutions.

Vikram Robotics Launches Affordable 'Vikram-1' Humanoid for Factory Tasks

Indian startup Vikram Robotics unveiled Vikram-1, a 1.6m humanoid priced at ₹2.5 lakhs (approx. $3,000), targeting indoor factory automation.

Research with immediate practical relevance

Stanford's SecureForge Improves Code Security via Automated Prompt Tuning

Researchers from Stanford introduced SecureForge, a pipeline that optimizes system prompts to significantly reduce security flaws in AI-generated code without sacrificing quality.

OpenAI Warns Upcoming Astra Model May Exceed Cybersecurity Threshold

OpenAI announced that its forthcoming Astra model might cross a critical cybersecurity threshold, triggering containment protocols. This is the first time OpenAI has explicitly named a model as presenting such a risk.