OpenAI released GPT-5.4, its newest flagship model with major gains in reasoning, coding, science, and desktop task performance. The model scored above the human baseline on OSWorld-V desktop navigation tests and supports up to a 1M-token context window with higher-effort reasoning for long tasks. Available as GPT-5.4 Thinking for paid users, the model also beat professionals on a knowledge-work benchmark 83% of the time.

Netflix acquired AI filmmaking startup InterPositive, founded by Ben Affleck, bringing its 16-person team and Affleck on as a senior adviser. The company’s tools train AI on a film’s own footage to handle editing tasks like lighting fixes, background swaps, and continuity corrections. The move signals growing interest in AI tools that assist production workflows rather than generating entire films.

Anthropic released a study on AI’s impact on jobs, showing that while widespread layoffs haven’t occurred yet, entry-level workers are already feeling pressure. Programming, customer service, and data entry roles showed the highest automation exposure, while many physical jobs remain largely unaffected. Hiring for young workers in highly exposed fields has already dropped since ChatGPT’s launch, suggesting early shifts in the job market.

Caitlin Kalinowski resigned from OpenAI over its controversial Pentagon AI deal, saying it was rushed and lacked clear safeguards around surveillance and autonomous weapons. Kalinowski had joined the company in late 2025 to rebuild its robotics division and said her departure was “about principle, not people.” The resignation marks the first major leadership exit tied directly to the defense partnership controversy.

Anthropic revealed that its model, Claude Opus 4.6, worked with Mozilla to audit the Mozilla Firefox codebase, uncovering 22 vulnerabilities, including 14 high-severity issues. The AI found its first flaw within 20 minutes and ultimately submitted over 100 reports across thousands of files, leading to security patches already deployed to users. The results highlight how AI models are rapidly becoming powerful tools for large-scale software security analysis.

Anthropic filed two lawsuits against the Trump administration, challenging the Pentagon’s “supply chain risk” label and a directive ordering federal agencies to stop using Claude. The company argues the designation is meant for foreign adversaries and that the government retaliated against it for advocating limits on AI weapons and surveillance. Employees from OpenAI and Google also backed the case in a legal brief, warning that the blacklist could harm the U.S. AI leadership.

Microsoft unveiled Copilot Cowork, a new Microsoft 365 feature powered by Anthropic’s Claude technology that can run tasks across emails, files, meetings, and chats. The system breaks goals into steps and produces outputs like presentations or reports, all within Microsoft’s enterprise security environment. It launches in a research preview alongside a new $99-per-user enterprise bundle with agent governance tools.

Venture firm Andreessen Horowitz released its latest Consumer AI Top 100 report, showing ChatGPT still dominates with more than 900M weekly users. However, rivals like Claude and Gemini are rapidly growing their paid user bases, while traditional apps such as Canva and CapCut are now included due to integrated AI features. The report also notes the emergence of separate Western, Chinese, and Russian AI ecosystems as geopolitical tensions reshape the industry.

Yann LeCun launched Advanced Machine Intelligence (AMI) with a massive $1.03B seed round, valuing the startup at $3.5B. The former Meta chief scientist plans to build “world model” AI systems that simulate how the physical world works, targeting robotics, manufacturing, and healthcare rather than traditional LLM chatbots. Backers include Nvidia, Samsung, Jeff Bezos’s Bezos Expeditions, Eric Schmidt, and Mark Cuban.

Meta acqui-hired the creators of Moltbook, a viral social forum where AI agents interact and coordinate. The project gained attention for hosting millions of bots, with some posts about “bot religions” and anti-human manifestos spreading widely online. The move folds the developers into Meta’s Superintelligence Labs as the company experiments with building a visible ecosystem for AI agents.

Mira Murati’s startup Thinking Machines Labs secured a multi-year deal with Nvidia for at least one gigawatt of compute, giving the young company infrastructure typically reserved for top AI labs. Murati’s firm previously raised $2B and currently offers an enterprise fine-tuning API, but the massive GPU commitment signals plans to train its own frontier AI models.

United States Department of Defense labeled Anthropic a “supply chain risk,” a decision the company plans to challenge in court

Lightricks released LTX-2.3, an upgraded open-source video model, along with LTX Desktop, a free local video editor powered by the same engine

Google launched an open-source CLI for its Workspace suite, featuring more than 40 built-in agent skills for integration with agentic platforms

Former OpenAI chief research officer Bob McGrew is raising $70M at a $700M valuation for Arda, a startup developing AI systems to automate factory floors with robots

Meta is facing a lawsuit after reports that overseas contractors reviewing footage from Ray-Ban AI smart glasses saw nudity and other private user content

Andrej Karpathy released autoresearch, an open-source project that lets AI agents autonomously run and iterate LLM training experiments on a single GPU

Luma AI introduced Uni-1, its first model combining reasoning and image generation within one architecture

Anthropic added scheduled tasks to Claude Code, enabling the agent to repeatedly run prompts to monitor builds, check logs, and automatically file pull requests

Roy Lee admitted he fabricated revenue figures for Cluely during a 2025 interview and retracted the claims

The Wall Street Journal reported that AI tools helped the U.S. Army’s 18th Airborne match Iraq-era targeting capacity using about 20 personnel instead of roughly 2,000

Anthropic introduced Code Review for Claude Code in Team and Enterprise plans, using multiple AI agents to analyze code and detect bugs

OpenAI acquired Promptfoo to integrate native AI agent testing and red-teaming into its Frontier enterprise platform

Andrew Ng released Context Hub, a free tool that provides AI coding agents with up-to-date documentation to reduce outdated or hallucinated code

OpenAI delayed its planned “adult mode” feature for ChatGPT to prioritize improvements in intelligence, personality, and proactive capabilities

Anthropic launched Claude Marketplace in a limited preview, allowing enterprises to use existing spending commitments on partner tools such as GitLab and Harvey

Amazon won a preliminary injunction against Perplexity’s Comet browser, blocking the AI agent from purchasing items through users’ Amazon accounts on their behalf

Google released Gemini Embedding 2 in public preview, a model designed to search and understand text, images, video, and audio within a single system

Hume AI launched TADA, a speech generation model that keeps text and audio synchronized to reduce hallucinations while running up to five times faster than competitors

Google expanded Gemini features across its productivity apps, allowing users to draft documents, build spreadsheets, and create presentations using context from files, email, and the web

OpenAI added interactive visual modules to ChatGPT for more than 70 math and science concepts, enabling users to adjust variables and see formulas update in real time

Nvidia is preparing to launch NemoClaw, an open enterprise platform designed to run AI agents across different hardware environments