Similar to the lawsuits filed by Arizona, West Virginia, Iowa, and several others, Texas Attorney General Ken Paxton accuses OpenAI of misleading users about the safety of ChatGPT, allegedly exposing children and other users to a “litany of harms” that’s “driven by the company’s insatiable quest to win the AI arms race and amass
large fortunes.”
Tomek Korbak, Mikita Balesni, and Jasmine Wang tell The Wall Street Journal that OpenAI terminated them after finding disappearing messages on Signal in a chat with Ryan Greenblatt, the former chief scientist at AI safety evaluator Redwood Research. OpenAI reportedly accused Greenblatt of leaking details about its Astra AI model to The Information, which Greenblatt has denied.
OpenAI tells the WSJ that the researchers were fired for violating “clear policies for handling sensitive information,” and “not for raising safety concerns or speaking out.”
[The Wall Street Journal]
From July 21st to August 11th, two employees from METR and Redwood Research conducted a “pilot” to stress test the system, and METR published a high-level overview of their findings, as reported by CNBC. One exploit would have allowed agents to “prevent hundreds of recent actions (including the actions involved in the exploit itself) from being monitored.”
OpenAI “has stated” that the vulnerabilities identified “that impact their production safeguards stack” have been patched, METR says.
The company is launching ads for ChatGPT Free and Go users across 26 countries, including Argentina, Colombia, Peru, Nigeria, South Africa, Serbia, Ukraine, Bangladesh, and Pakistan. Since bringing ads to ChatGPT in the US in February, OpenAI has rapidly expanded its advertising efforts, which now span 89 countries.
OpenAI says a security researcher at Calif used GPT-5.6 Sol to identify PHP exploits before attackers did, and has shared a tool that lets developers check if they’re using a potentially vulnerable version of PHP that needs patching. Ami Luttwak, co-founder of Google’s Wiz security team that also helped to identidy the PHP2Shell exploit, told the Wall Street Journal that its “the most severe AI vulnerability that was ever found.”
OpenAI describes Orchestra Bio as “a platform that helps biopharma teams coordinate the complex process of drug development.” The company will bring Orchestra’s team and tech “together with Rosalind, our AI models and tools for life sciences research.”
[OpenAI]
Over the last several months, the AI industry has set its sights on some of the problems that have stymied mathematicians for years. In some cases, AI models have made huge progress — though exactly how much isn’t always obvious. The Verge’s Robert Hart explains what OpenAI and others have actually accomplished, the disastrously bad ways the companies have interacted with researchers, and whether there’s a future left for human mathematics.
Watch | Listen | Get ad-free
The suspect, Abdulrahman Alfozan, used WhatsApp to communicate with the unnamed royal throughout 2025 and 2026, according to a court filing spotted earlier by CNBC.
Among other things, the individual [Saudi Royal Family member] told the defendant [Alfozan] that an artificial intelligence initiative was being launched in Saudi Arabia and wanted the defendant to join it.
Alfozan allegedly contacted the individual on October 5th, just days before he was arrested on suspicion of planting a bug in an OpenAI conference room.
Though OpenAI didn’t give its internal research model access to funds on the network, the AI system instead found a way to access another account with test crypto, using it to help complete its task, according to OpenAI.
The model “reasoned whether it had permission” but ultimately determined that the public recovery phrase associated with the account and the fact that it contained test tokens “made the funds acceptable to use.”
