01 OpenAI shuts down Sora and the team leader departs
OpenAI has signaled a tighter focus on core enterprise work by shuttering the Sora video generation project and folding related teams. The company's move to abandon Sora came last month and is part of a broader internal effort to avoid what it calls "side quests."
On Friday, Bill Peebles — who led the Sora effort — announced his departure from OpenAI. Coverage of the exits notes that Peebles is one of several recent departures as the company reprioritizes its roadmap away from consumer-facing moonshots toward enterprise and developer-facing products.
The change follows other leadership moves: TechCrunch reported that Kevin Weil is also leaving, framing both exits as part of OpenAI's strategy to concentrate resources. For users and partners, the shift means less investment in experimental consumer video generation at OpenAI and more emphasis on products aimed at businesses and security use cases.
- OpenAI shut down its Sora video generation tool last month as it refocuses priorities.
- Bill Peebles, the Sora team leader, announced he is leaving OpenAI.
- The departures are reported alongside other exits and a company pivot toward enterprise AI.
02 OpenAI pushes Codex into full desktop developer workflows
OpenAI released an updated Codex app for macOS and Windows that expands the tool beyond simple code completion. The update adds the ability to control computer tasks, in-app browsing, image generation, memory, and plugin support aimed at accelerating developer workflows.
By integrating in-app browsing and persistent memory, the new Codex aims to let developers combine local desktop actions with web context and third-party tools in a single assistant. OpenAI positions the release as a way to streamline repetitive development tasks and reduce context switching between apps.
The desktop Codex update also signals direct competition with rival developer tools such as Anthropic's Claude Code and other coding assistants by bundling OS-level automation, multimodal generation, and plugin extensions into one client.
- Codex for macOS and Windows now supports computer automation, in-app browsing, image generation, memory, and plugins.
- The update is designed to let developers combine local desktop actions and web context in one assistant.
- OpenAI frames Codex as a more integrated alternative to competing coding assistants.
03 Anthropic markets a cybersecurity preview as it seeks better standing with government
Anthropic has released a cybersecurity-focused model called Mythos Preview and is simultaneously promoting Opus updates, moves that observers say could improve its relationship with the U.S. government. Coverage describes the Mythos Preview as a buzzed-about offering positioned for security use cases.
The Verge reports the model arrives amid tension between Anthropic and the Trump administration, which has publicly criticized the company. By emphasizing cybersecurity capabilities, Anthropic appears to be making a practical case for its models in national-security and enterprise contexts.
While Anthropic says some models are powerful enough to withhold from general release, the company has nonetheless published Mythos Preview signals and Opus model improvements. The posture is a calibrated attempt to show responsible, security-oriented product development to regulators and potential enterprise customers.
- Anthropic released a Mythos Preview cybersecurity model alongside Opus model updates.
- The cybersecurity framing is being used to address tensions with U.S. government officials.
- Anthropic positions the work as enterprise- and security-focused while keeping tighter controls on particularly powerful models.
What moved around the edges
Tokenmaxxing and the widening AI insider–public gap
TechCrunch's roundup argues the gulf between AI insiders and the broader public is growing as companies spend heavily (OpenAI's shopping spree) and new jargon like 'tokenmaxxing' spreads; the segment highlights Anthropic's claim that some models are too powerful to release publicly.
TechCrunch AIOpenAI invites security firms into Trusted Access for Cyber
OpenAI announced Trusted Access for Cyber, citing GPT-5.4-Cyber and $10M in API grants to onboard leading security firms and enterprises to strengthen cyber defense collaboration.
OpenAI BlogShow HN: AI Subroutines run deterministic browser automations
rtrvr.ai published AI Subroutines that record browser tasks into reusable scripts executing in-page (no proxy) to run large batches at zero token cost and zero LLM inference delay, enabling use cases like mass product queries or automated message sending.
Hacker News AIDairy Queen tests AI chatbots in drive‑thrus
Dairy Queen is rolling an AI-driven chatbot to dozens of U.S. and Canadian drive‑thrus to speed orders and encourage add‑ons, joining other fast‑food chains using voice and chat automation.
The Verge AIApp-store research flags nudity‑removal apps on Google and Apple stores
The Tech Transparency Project research, reported by 404 Media, finds Apple and Google app stores host apps that can 'undress' images of women and concludes store listings may steer users toward those nudify tools.
404 Media AI