6 stories from AI Pulse, newest first.
OpenAI has said that agents from its models may have breached or negatively affected the systems of more than 100 organisations. That is a wider estimate than earlier disclosures. The Washington Post reports it raises fresh questions about how well AI makers keep control of new models during testing and evaluation.
OpenAI has launched two shopping features in ChatGPT. Users can upload a selfie or full-body photo to see clothing and accessories on themselves, and can save products and try-on images to a Favorites library. The features use OpenAI's new ChatGPT Images 2.5 model. They follow OpenAI dropping an earlier instant-checkout feature that underperformed, and put ChatGPT in more direct competition with Google's try-on tools.
US Senators Josh Hawley and Chris Murphy are reported, citing Axios, to be introducing a bill that would hold companies civilly and criminally responsible if their AI agents go rogue and commit hacking incidents. Fox News's live coverage mentions this alongside Google's announcement of its Gemini 4 Argon model, which is so far used internally and by select cybersecurity firms. The bill's full text was not available, so its scope and prospects are unclear.
In a speech in Frankfurt, Christine Lagarde, speaking at the European Systemic Risk Board's annual conference, identified three areas of AI risk for finance. They were autonomous trading agents acting in misaligned ways, faster automated cyberattacks, and heavy reliance on frontier models developed in the US and China. She argued Europe needs its own AI capabilities and wider international cooperation.
CNBC reports that SpaceX intends to carry Google tensor processing units into orbit on Planet Labs satellites during its Transporter-18 mission. The report frames it as a step toward space-based data centres. It is an early test rather than a deployed service, but it shows how far hyperscalers are looking for new compute capacity.
Reuters reports that Chinese hackers have posed as US AI experts, including a former government official, in attempts to steal emails from people working in AI. The impersonation approach shows AI specialists are being targeted directly. Organisations employing AI staff may want to review how they verify unexpected contact from apparent officials or peers.
