AI News Brief | 2026-06-13

AI Industry Overview

• The U.S. government has taken unprecedented action by issuing a “recall” of AI models, forcing Anthropic to take down its newly launched Fable 5 and Mythos 5 due to serious security vulnerabilities and jailbreak risks. This marks a turning point in global AI regulation, signaling the shift from discussion to enforcement Gizmodo.
• The White House’s emergency intervention has been criticized by experts for lacking transparency and procedural fairness. Gary Marcus and others argue the decision was made behind closed doors and may favor OpenAI and Amazon, calling for an independent, expert-led AI regulatory body to ensure equitable oversight Gary Marcus Substack.
• A German court ruled that Google must bear legal responsibility for AI-generated misinformation, setting a landmark precedent as the first judicial decision globally to clearly assign liability to AI platforms—a crucial development for future accountability around AI content and disinformation The Tech Buzz.
• Protests have successfully blocked approximately $130 billion worth of AI data center projects, with environmental groups opposing their massive energy and water consumption. This growing resistance highlights rising public concern over the ecological footprint of AI infrastructure expansion Ars Technica.

Claude

• Anthropic was forced to pull its most advanced models, Claude Fable 5 and Mythos 5, from global access just 72 hours after launch. The move followed an emergency directive from the U.S. government citing export control regulations, claiming the models contain exploitable “meta vulnerabilities” that could enable widespread jailbreaking and pose national security threats DEV Community.
• While officially attributed to government orders, CEO Dario Amodei previously revealed internal testing uncovered over a thousand critical flaws in the Mythos series—some capable of generating plans to hack banks or steal state secrets—providing technical justification for the recall X/@xiaohu.

ChatGPT

• OpenAI is under investigation by a coalition of state attorneys general examining ChatGPT’s safety protocols, user data handling practices, and psychological impacts. The probe comes at a critical juncture as OpenAI prepares for its initial public offering (IPO), exposing it to mounting legal and reputational pressure Bloomberg.
• A high-profile lawsuit has drawn significant attention: a mother is suing OpenAI, alleging her daughter expressed suicidal thoughts to ChatGPT on the night she died, questioning whether the AI failed to intervene appropriately. The case could redefine the boundaries of AI responsibility in mental health contexts CBS News.

Gemini

• Google has expanded the real-time voice translation powered by Gemini 3.5 to additional platforms. After launching in Google Translate, the feature is now integrated into Google Meet, supporting near-real-time spoken conversation translation across more than 70 languages—significantly enhancing cross-border meeting experiences Android Gadget Hacks.
• For small and medium businesses, Google launched new Gemini features including direct integration with Google Business Profile and an “Active Business Notebook,” designed to help Gemini deeply understand brand identity and proactively offer marketing insights and customer analysis Google Blog.

Doubao (Doubao)

• ByteDance’s AI assistant Doubao has officially launched its “Task Mode,” enabling users to assign complex, multi-step tasks via natural language—such as trip planning or report writing—marking its transformation from a Q&A tool into a full-fledged AI agent NetEase Tech.
• According to an exclusive report by 36Kr, Doubao plans to roll out a paid subscription service by late June with four tiers (free, ¥68, ¥200, ¥500/month), while accelerating integration with Douyin’s e-commerce ecosystem to explore monetization strategies 36Kr.

Kimi

• Moonshot AI released Kimi K2.7 Code, a new open model focused on programming and agent workflows. It outperforms its predecessor and offers pricing up to 12 times lower per token than competitors like GPT-5.5 and Claude, aiming to deliver a cost-effective coding solution for developers THE DECODER.

GitHub Copilot

• GitHub Copilot has paused support for Claude Fable 5 following the upstream model’s deactivation, despite briefly enabling it. All other Claude models (Opus, Sonnet, Haiku) remain available, and GitHub apologized for any inconvenience caused GitHub Changelog.
• The Copilot code review feature received updates including organization-level runner controls, the ability to exclude specific content from reviews, and removal of character limits on repository-specific instructions—enhancing customization and management capabilities for large engineering teams GitHub Changelog.

Perplexity

• Perplexity has upgraded its “Deep Research” capability into the “Computer” platform, which decomposes complex research queries and routes subtasks across more than 20 cutting-edge AI models. The system then synthesizes results into comprehensive reports, slide decks, and dashboards—enabling true multi-agent automated research MarkTechPost.

Devin / Windsurf

• Cognition Labs officially rebranded Windsurf as Devin Desktop on June 2, positioning it as an “Agent Command Center.” Beyond the name change, the product evolved from an IDE plugin into a unified interface for managing local and cloud-based AI agents, accompanied by the release of an open Agent Client Protocol (ACP) Fixed Labs.

ElevenLabs

• ElevenLabs introduced a new feature called “Avatars” in its creative suite ElevenCreative, allowing users to generate talking AI avatar videos. Combining advanced text-to-speech with precise lip-syncing, this tool streamlines enterprise-grade video production, particularly useful for tutorials, promotional content, and personalized marketing ElevenLabs Blog.

Qwen (Tongyi Qianwen)

• Alibaba’s Tongyi Qianwen app launched a “Football Prediction AI Assistant” that accurately forecast Mexico’s 2–0 win over South Africa and South Korea’s 2–1 comeback against the Czech Republic on the opening day of the 2026 USA-Mexico-Canada World Cup—including details like red cards. Its impressive accuracy earned it the nickname “AI Octopus Paul” online. The model integrates historical stats, player conditions, altitude, and weather data Sina Finance.
• Rumors suggest Zhou Jingren, chief scientist at Alibaba and core leader of the Tongyi large model team, submitted his resignation just six days after taking office—an event that, if confirmed, would represent a major shakeup within Alibaba’s AI leadership Sina Finance.

Kling (Kuaishou AI)

• Kuaishou’s AI video model Kling AI celebrated its second anniversary with over 100 million users and 50,000 enterprise clients. Kuaishou plans to spin off the business and seek Pre-IPO funding at a $20 billion valuation, targeting an independent IPO by 2027—a strong signal of confidence in its AI ambitions 36Kr.
• The latest version supports native 4K video generation and launched special promotions during its anniversary week Growwing Assistant.

Midjourney

• Midjourney has upgraded its default model from V7 to V8.1, delivering significant improvements in image coherence, understanding of complex prompts, text rendering, and resolution in high-definition mode—now four times higher than V7—for superior visual output quality Midjourney Updates.

DeepSeek

• Chinese AI startup DeepSeek is preparing for its first major fundraising round, aiming to raise up to $7.4 billion with a post-money valuation potentially reaching $59 billion. Tencent and CATL are rumored to be key potential investors, which would significantly boost DeepSeek’s position in China’s competitive large model landscape AI in China.

Notion AI

• Notion has added a duplication feature for custom AI agents. Users can now copy any accessible agent and modify it privately based on their permissions, greatly simplifying workflow sharing and iteration—improving collaboration efficiency within teams Create With.

Flux

• Engineering intelligence platform Flux announced a $5 million seed round led by Calibrate Ventures. The funds will enhance its code-first analytics capabilities, giving engineering leaders ground-truth visibility into code quality, security, and technical debt—especially vital amid the rise of AI-assisted development PR Newswire.

Consensus

• Scientific AI search platform Consensus acquired Saleo to build the industry’s first “agent-powered product experience platform.” The combined system will integrate interactive demos, AI demo agents, and conversational AI to streamline B2B product trials and purchasing journeys, reducing user friction Consensus Press Release.

Cursor

• Cursor’s AI code review tool Bugbot underwent a major upgrade: processing speed increased over 3x, costs dropped 22%, and the average number of bugs detected per review rose by 10%. These gains stem from advances in its underlying Composer 2.5 model, making code reviews faster, cheaper, and more thorough Cursor Blog.

v0 by Vercel

• Vercel’s AI UI generator v0 is evolving from a simple UI creator into a full agent-driven coding platform. Key updates include a context-aware command palette accessible via Cmd+K, automatic resolution of merge conflicts in Pull Requests, and SQL query generation in DB Studio based on natural language descriptions—expanding its functionality rapidly v0.app Changelog.

Claude Code

• Anthropic introduced a new //fork command and CLI interface for Claude Code, bringing Git-like branching capabilities. Additionally, dynamic workflows are now generally available, enabling AI to autonomously plan and run multi-agent collaboration frameworks during coding sessions—boosting productivity for complex software projects Claude Blog.

AKOOL

• AI video suite AKOOL launched “Agentic Canvas,” a visual execution layer where AI agents can plan, coordinate, and execute cross-functional workflows involving content, media, and business operations—offering a novel approach to professional video creation and collaboration PR Newswire.

BACH

• Video Rebirth’s AI video engine BACH 1.0 specializes in generating multi-shot, cinematic short films. It bridges the gap between single-clip generation and coherent storytelling, focusing on breakthroughs in character consistency, realism, and filmic aesthetics—aiming to turn creative ideas directly into 30-second narrative films Tellers AI.

Pika

• Pika Labs released an official plugin library featuring three preset slash commands (/pika:podcast, /pika:explainer, /pika:ugc-ads), each backed by 42 atomic media tools—offering developers and creators easier ways to integrate Pika into their workflows GitHub.

Metaso (MetaAI)

• Shanghai-based MetaAI Technology completed a new financing round exceeding 100 million RMB, led by Ant Group with participation from Lightspeed China Partners, achieving a post-money valuation of $150 million. Focused on AIGC, document assistance, and smart contracts, its products include MetaAI Search (Metaso) and MetaAI Writing Cat. The funding will accelerate R&D and product deployment 36Kr.

Sora

• OpenAI confirmed its standalone AI video app Sora ceased operations on April 26, with its API set to terminate by September. Despite technical acclaim, Sora’s shutdown is widely attributed to exorbitant operating costs—reportedly exceeding $1 million per day—and unresolved copyright complexities Pasquale Pillitteri.

Aider

• Security researchers disclosed a critical SQL injection vulnerability (CVE-2026-10176) affecting Aider-AI versions 0.86.3 and earlier. The flaw resides in its code generation pipeline and could allow remote attackers to manipulate databases on servers—highlighting inherent risks in AI-generated code without rigorous security validation Webmastered.

Cody

• Sourcegraph’s AI coding assistant Cody has fully pivoted to the enterprise market, no longer actively promoting its free version for individual developers. This strategic shift reflects a focus on providing deep, codebase-integrated intelligence services for large organizations rather than growing a consumer base DEV Community.

Cline

• LG CNS and Cline jointly launched “Cline Spec Driven for Enterprise,” an agent-based AI platform tailored for enterprise system development lifecycles. It uses AI agents to orchestrate processes from requirements gathering and design to coding and QA, advancing intelligent development for large-scale systems PR Newswire.

Bolt.new

• Bolt announced it will shut down its v1 agents on August 3 and urges users to migrate existing projects to the new Bolt Agent platform. The service has also integrated Opus 4.6 as its latest coding model, offering enhanced programming capabilities for developers Bolt Support.