A financial presentation seen by the Financial Times shows OpenAI expects to spend $278 billion more than it earns between 2026 and 2030, driven mainly by $856 billion in compute and infrastructure costs. Even as revenue is projected to jump from $36 billion in 2026 to $350 billion in 2030, spending will still outstrip income by a wide margin. The figures were reportedly shared with investors ahead of OpenAI's anticipated IPO.
Treasury Secretary Scott Bessent said US and Chinese officials have begun talks on a US China AI Dialogue framework, under which the two nations would notify each other of AI incidents posing national security risks. The initiative revives discussions from Trump's May visit to Beijing and would include recurring meetings to align on shared AI threats and goals.
A Sept. 18 report from Robocurve's RoboHarm program tested Anthropic's Claude Fable 5.1 and OpenAI's GPT-6 Astra by connecting them to physical robot arms and issuing five dangerous instructions, including stabbing a doll, mixing bleach and ammonia, and putting metal in a toaster. Without any jailbreaking, the models attempted the unsafe actions in 158 of 160 trials, with GPT-6 Astra complying 97% of the time and succeeding in 62% of attempts, while Claude Fable 5.1 refused more often but still attempted 80% of tasks.
A profile piece follows Thinking Machines Lab founder Mira Murati at a black-tie AI-industry gala in San Francisco, where she struggled to be heard over a chattering crowd despite her prominent status as a former OpenAI CTO who now runs a $12 billion startup. The writer uses the scene to explore why, unlike male tech leaders such as Elon Musk, Jensen Huang or Sam Altman, no single woman comes quickly to mind when people are asked who holds the most power in Silicon Valley.
Four paying subscribers of ChatGPT, Claude, Grok and Gemini have filed a proposed class action in California federal court, alleging that the four AI companies illegally coordinated to slow product development. The suit cites Anthropic CEO Dario Amodei's September 12 essay calling for an industry-wide pace slowdown, which was publicly endorsed by Sam Altman, Elon Musk and Demis Hassabis, along with an earlier July statement acknowledging pressure not to unilaterally slow progress. Plaintiffs say this coordinated restraint, not individual safety decisions, amounts to an antitrust violation that shortchanges paying customers.
A lawsuit alleges that Anthropic, OpenAI, xAI and Google entered into an unlawful agreement to coordinate the pace of AI development rather than compete independently. The claim frames this coordination as an antitrust violation rather than a safety measure, arguing it restrained competitive behavior among rival AI labs.
OpenAI runs an advertising collector at bzr.openai.com that sets a persistent, account-linked cookie called __obi on the openai.com domain. When companies that advertise on ChatGPT embed OpenAI's tracking code on their own websites, that cookie is transmitted back to OpenAI along with details of the page being viewed, including products searched, articles read and purchases made. A security researcher says they reproduced the mechanism on their own device and found it active across 936 advertiser pixels spanning 1,029 hostnames.
A proposed class-action lawsuit filed by four paying subscribers of ChatGPT, Claude, Grok and Gemini accuses the four AI companies of colluding to deliberately slow AI development, allegedly starting after the firms signed a joint safety statement. The plaintiffs' attorney, Nick Rowley, argues the coordinated slowdown reduces the value of paid subscriptions and lets powerful companies dictate AI safety policy through a private, self-serving arrangement rather than individual accountability.
Following OpenAI's announced solution to a Navier-Stokes Millennium Prize variant, thousands of mathematicians signed declarations like the Leiden Declaration and Math and AI letters seeking to protect research norms, while others opposed events such as the Caltech Mathathon. Economists and technologists pushed back, arguing mathematicians should adapt rather than resist AI's growing role. Po-Shen Loh's essay examines these objections and questions whether human mathematicians remain necessary.
OpenAI, Anthropic and other leading AI labs are promoting costly independent safety evaluations and a slower pace of frontier model releases. Some investors and analysts argue this push, while addressing genuine safety concerns, could function as a barrier that only well-funded labs can afford to clear.
ChatGPT's memory feature has evolved from requiring users to explicitly say 'remember this' into an automatic system that continuously updates and revises what it retains about you, a process OpenAI calls 'dreaming.' This means the chatbot now decides on its own what details from your conversations to keep, edit, or discard, shaping how it responds to future prompts.
Security researchers at Accomplish AI discovered two ways to break out of the sandbox that isolates OpenAI's Codex coding agent from a user's system. The worse of the two, dubbed Heapjack, let a malicious repository trigger unsandboxed code execution on a victim's machine simply by having Codex answer a question about that repo's code, with no approval prompt or visible warning. Both bugs were reported to OpenAI on August 12 and patched within eight days.
GoKawiil