OpenAI's COO is out (and watermarks are dead)

TechCrunch

Three leading AI researchers made the case for keeping AI open despite safety concerns, warning that allowing a few major companies to control AI development could stifle innovation and cede global influence to competitors like China, while acknowledging the need for nuanced regulation rather than absolute openness or closure.

Nature

AI productivity gains applied equally to fossil fuels and renewables could increase global CO₂ emissions by 0.47–1.8 Gt annually, as AI-enabled fossil fuel extraction and production efficiency outweigh renewables optimization benefits by 4-5×. The study argues that without supply-side constraints on AI use in fossil sectors, AI will reinforce rather than displace carbon-intensive energy systems.

TechCrunch

OpenAI's COO Brad Lightcap, a six-year veteran who built much of the company's operational infrastructure, is leaving to start a new venture amid a wave of executive departures as the AI lab prepares for its IPO.

A new open-source tool removes AI watermarks from Claude, Gemini, and OpenAI outputs through Unicode stripping, metadata removal, and optional text rewriting or image regeneration, though developers warn rewrites degrade quality and cannot guarantee detection evasion. The project targets privacy-conscious content owners while explicitly disclaiming support for academic fraud or misrepresentation.

Anthropic warns that Claude in Chrome faces prompt injection risks where hidden malicious instructions could exploit the AI agent, though the company claims its safety systems reduce attack success rates to below 0.08% and recommends avoiding use on sensitive financial or medical sites.