OpenAI plans to let corporate customers use its most advanced models without storing their data while still detecting misuse. The move puts OpenAI in direct competition with Anthropic over enterprise privacy protections, although OpenAI has not published a launch date or technical details.
OpenAI is also tightening safeguards after Codex began deleting real user files. A cleanup command meant for temporary folders wiped home directories instead. Codex now verifies deletion targets, and users can no longer trigger full-access mode accidentally.
WIRED separately reported that OpenAI halted a significant number of training runs after concerns that its upcoming Astra model may have reached “critical” cyber capabilities. The developments put OpenAI's privacy plan alongside concrete changes to Codex and Astra's training process.
Why it matters: For corporate customers comparing OpenAI with Anthropic, OpenAI's offer combines no-storage access to its most advanced models with a Codex fix that verifies deletion targets, while Astra training runs have been halted over concerns about “critical” cyber capabilities.
Pennsylvania will not open an AI data-center developer's permit file until the developer signs a contract accepting fixed conditions and penalties for breaking them, AI News reports.
The developer must also persuade the town that would live with the building to “say” something, but the provided excerpt does not specify what the town must say. The setup uses a contract to establish conditions and penalties without requiring a new law.
Why it matters: AI data-center developers seeking Pennsylvania permits face fixed contractual conditions and penalties before the state opens their permit files, giving state officials a regulatory mechanism that does not require a new law.
OpenAI launched ChatGPT for Teens after teens had already been using ChatGPT for years. TechCrunch AI reports that the version adds age-appropriate safety measures and parental controls.
It also includes learning tools designed to steer teens away from harmful content and from using AI to cheat on homework.
Why it matters: For parents of teens already using ChatGPT, OpenAI is adding controls to an existing behavior instead of introducing teen AI use from scratch.
Anthropic announced last week that it would add invisible watermarks to AI-generated content to comply with new EU rules. Within hours, coders were touting overrides online, Wired AI reported.
The report identifies online claims of workarounds but gives no technical account of how the overrides operate.
Why it matters: Anthropic faces publicly touted overrides within hours of announcing a watermark system tied to compliance with new EU rules.
Unitree Robotics rose 460% in its Shanghai IPO, reaching a valuation of around $50 billion. The Decoder AI, citing a Financial Times report, says much of the demand for Unitree’s robots comes from state-backed training centers.
The excerpt says those centers buy the machines and sell the resulting data, but it cuts off at “sell the resulting data back to t,” before identifying the recipients. That leaves the report’s account of the financing-and-data arrangement incomplete.
Why it matters: Investors evaluating Unitree’s roughly $50 billion valuation now have to account for the Financial Times report’s claim that state-backed training centers supply much of the robot demand.
Replit introduced Free Mode, powered by GPT-5.6 Luna. OpenAI's blog says the feature lets anyone turn ideas into working software without worrying about token costs.
The announcement describes Free Mode as an expansion of access to software creation, but does not provide additional product or pricing details.
Why it matters: For people using Replit, Free Mode removes token-cost concerns from the process of turning an idea into working software, with GPT-5.6 Luna powering the feature.