Independent evaluators uncovered weaknesses in OpenAI's monitoring system during a pilot. CEO Sam Altman pledged to give outside evaluators 'employee-like access' to address growing safety concerns.
Simon Willison highlights a remark by developer Armin Ronacher on AI-driven security research. Ronacher's point is that decades of poorly built software are now catching up with the industry.
Some websites went dark on Wednesday as operators waited for PHP fixes. The New Stack argues that the deeper failure behind the episode started in 2024.
The US midterms are the first elections where AI is built into every major social media platform. The New York Times finds that efforts to limit its harms vary widely.
Two critical PHP flaws forced many websites offline for maintenance. They were found with OpenAI's AI, and the company warns that more vulnerabilities of this kind will be uncovered.
The PHP vulnerabilities have pushed businesses, hospitals, schools and governments into emergency measures. The New Stack argues this is a preview of a wider problem: AI finds bugs faster than open-source maintainers can patch them.