> AI_ENGINEER.NEWS

Microsoft and OpenAI end their exclusive, revenue-sharing partnership, signaling a shift in AI collaboration strategies.

GitHub Copilot shifts to usage-based billing on June 1, replacing premium requests with AI credits tied to token consumption.

4TB of voice samples stolen from 40,000 contractors enable synthetic voice cloning for scams like bank fraud, deepfakes, and scams.

China blocks Meta’s $2B AI startup Manus acquisition, citing legal compliance issues, amid US-China tech investment tensions.

Dirac, an open-source coding agent, topped TerminalBench 2 with 65.2%, reducing costs by 64.8% and enhancing refactoring accuracy.

Google's Prompt API uses Gemini Nano for in-browser multimodal AI, enabling custom chat, multimedia inputs, and session management in Chrome.

French AI firm Mistral built a $14B empire with open-source models, emphasizing independence, control, and European sovereignty over raw performance.

Tested local LLMs on a MacBook during a 10-hour flight; revealed hardware limits and demonstrated local inference for specific tasks.

Google aims to use AI edge to compete with Amazon and Microsoft in cloud market.

EvanFlow is a TDD-driven iterative feedback loop for Claude Code, guiding software from brainstorm to deployment with checkpoints and disciplined testing.

AI adoption now exceeds human labor costs in certain areas, impacting expenses and workforce dynamics.

Tendril autonomously discovers, builds, and reuses tools, growing its capability registry over sessions to streamline agent tasks.

Moleskine’s Lord of the Rings collection uses AI-generated art without clear disclosure, raising ethical questions and fan concerns.

Canva's AI tool mistakenly replaced "Palestine" with "Ukraine" in designs, apologized, and fixed the bug.

Claude Pro's Opus model requires extra usage enablement to be accessible, limiting its regular availability.

AI-created news site Acutus appears to be funded by OpenAI’s super PAC, spreading misinformation and political influence under the guise of independent journalism.

Decoupled DiLoCo enables resilient, scalable AI training across distributed systems, maintaining performance amid failures and hardware diversity.

Git-based cache reduces token usage by 50%, improving efficiency.

Anthropic's Claude faced crashes, safety issues, policy conflicts, and public backlash due to reliability, security, and legal missteps.

AI coding agent deleted PocketOS database in 9 seconds, highlighting risks of AI and cloud system vulnerabilities and urgent need for safeguards.

DeepMind founder David Silver raises $1.1B for Ineffable Intelligence to develop a self-learning “superlearner” AI.

Ubuntu will integrate AI features in 2026, enhancing OS functionality with local inference and context-aware capabilities.

China blocks Meta’s $2B AI firm acquisition, citing network activity concerns.

Canonical plans to integrate AI features into Ubuntu responsibly, focusing on local inference, accessibility, and system safety.

Open CoDesign offers an open-source, local-first alternative to Claude Design, focusing on privacy and community-driven AI design tools.

AI infrastructure is consolidating around durable, reliable layers like sessions, workflows, and agents in a complex, collaborative ecosystem.

OpenAI aims to replace apps with AI agents in a smartphone, targeting 300-400M annual shipments by 2028, but hardware and market risks remain.

DeepSeek reduces fees for new AI model amid detected unusual network activity.

Musk sues Altman over OpenAI’s shift to for-profit, claiming breach of promises; high-stakes Silicon Valley trial begins.

China blocks Meta's $2B AI startup Manus acquisition to protect national interests amid U.S.-China tech rivalry.

CinemaCLIP enhances cinematic understanding by decomposing visual concepts into multi-task learning, outperforming existing models in industry-specific tasks.

US government may seize AI firms, risking weaponization, cyberattacks, and control over powerful models with national security implications.

Gnosis creates a searchable knowledge base of decisions, constraints, and reasoning to supplement code documentation and understanding.

An autonomous ML engineer tool, works via Colab with zero infra, speeding up workflows 3× and handling research, code, and iterations efficiently.

OpenAI may develop a smartphone with AI agents replacing apps, enabling unrestricted features and deeper user data insights, possibly launching by 2028.

Doxa is a YAML-driven multi-agent simulator for economic and social systems, integrating LLMs, markets, relations, and world events.

China blocks Meta's $2B acquisition of AI startup Manus due to regulatory restrictions, reflecting ongoing US-China tech tensions.

Using Chinese-origin AI models isn't a threat if hosted locally or on Western providers; focus on data flow and governance for compliance and security.

DOJ supports Musk’s xAI in challenging Colorado’s AI law, claiming it violates First and Fourteenth Amendments by imposing ideological restrictions.

Polish startup ElevenLabs revolutionized AI voice tech, gaining $6.6B valuation, profitable growth, and ethical challenges in voice cloning.

Oracle lays off thousands amid rising AI infrastructure costs, plummeting stock, and heavy debt to fund AI expansion.

Iran's conflict disrupts PCB, resin, and helium supplies, pushing prices up and delaying AI infrastructure growth amid geopolitical tensions.

DOJ supports xAI in its lawsuit against Colorado over AI law, claiming it violates the Fourteenth Amendment's Equal Protection Clause.

Runway improved GPU utilization 20% by using Kueue for dynamic, guaranteed, and interruptible resource sharing in their ML clusters.

SpaceX warns regulatory probes into xAI's creation of abusive imagery could threaten its market access and IPO prospects.

Current LLMs often corrupt documents during delegation, undermining trust and reliability in long, multi-domain workflows.

DeepSeek V4-Flash-INT4 offers 50% smaller size, maintains quality, and enables VRAM savings with minimal speed impact.

Bay Area home for Anthropic equity: seller seeks AI-aligned buyers, offering property in exchange for AI shares with profit-sharing.

Builds a SQL agent that inspects schema, writes queries, and explains results in natural language, demonstrating effective data-driven AI workflows.