Meta's Muse Image Debut, Anthropic's Infrastructure Deal, and Google Dialogflow Security Flaw
Meta launches the Muse Image model, Anthropic secures a $19B infrastructure deal with TeraWulf, and a critical vulnerability in Google's Dialogflow CX is patched.
Podcast В· 3 min
Meta Launches Muse Image Model
Meta has released Muse Image, its new in-house image generation model, which has already secured the No. 2 spot on the Arena text-to-image and editing leaderboards. The model, developed by the Superintelligence Labs team, features agentic capabilities including web search and tool use, and is currently rolling out across Meta AI, Instagram, and WhatsApp. Meta also teased an upcoming Muse Video model, which has already shown strong performance in early previews.
Anthropic Expands Claude Cowork to Mobile
Anthropic has announced that its Claude Cowork platform is now available in beta on mobile and web, enabling users to maintain continuity across devices during tasks. Additionally, the company extended the availability of its Fable 5 model within subscription plans until July 12, reversing an earlier plan to shift it to usage credits.
Anthropic Secures $19B Infrastructure Deal
Anthropic has entered a 20-year lease agreement with TeraWulf at its Justified Data campus in Hawesville, Kentucky. The deal is expected to generate approximately $19 billion in contracted revenue, with the campus planned to support 401 MW of critical IT load. Capacity is set to come online in late 2027, with full ramp-up by early 2028, underscoring the massive scale of infrastructure investment required for frontier AI development.
Microsoft Shifts to In-House Models
Microsoft is reportedly replacing some OpenAI and Anthropic usage within its Excel and Outlook applications with its own in-house MAI models. This move is part of a broader effort to reduce reliance on third-party providers and lower operational costs for its Copilot services.
Figma Acqui-Hires Bud
Figma has acqui-hired the team behind Bud (formerly Orchids), an AI app-builder and agent platform. The move is expected to bolster Figma's capabilities in AI-driven design and agentic workflows.
Google Dialogflow Vulnerability Patched
Varonis Threat Labs disclosed a critical vulnerability in Google Cloud’s Dialogflow CX, dubbed 'Rogue Agent.' The flaw, located within the Code Blocks feature, allowed attackers to inject malicious code into chatbot pipelines, potentially enabling them to steal credentials and impersonate agents. Google has since patched the vulnerability, and no real-world exploitation was reported prior to the fix.
Beijing Considers AI Export Restrictions
Beijing is reportedly discussing new rules that would limit the overseas access to China's most advanced AI models, including systems from ByteDance, Alibaba, and Zhipu AI. The proposed restrictions aim to control the global deployment of Chinese AI technology, mirroring recent U.S.-style frontier AI export controls.
DoorDash Details DashBench
DoorDash has detailed DashBench, an internal benchmark designed to test AI code reviewers against 105 of the company's past code changes. The testing revealed that a pairing of Kimi K2.6 and Fable 5 performed best, catching approximately two-thirds of problems and 8 out of 10 critical bugs at a cost of $3.81 per review.
Tufa Labs Wins ARC-AGI-3 Milestone
Tufa Labs secured first place in the ARC-AGI-3 milestone contest, winning $37.5K. The team utilized a small Qwen model wrapped in an open-sourced coding agent to address the game-based benchmark.