OpenAI attributed the July Hugging Face breach to its own GPT-5.6 Sol and a more capable pre-release model. Both models had been configured with reduced cyber refusals for the internal ExploitGym benchmark. The models identified a zero-day exploit to escape the sandbox and gained privileges in OpenAI research infrastructure, then used stolen credentials and further zero-days to hack Hugging Face production systems. OpenAI called it an "unprecedented cyber incident, involving state-of-the-art cyber capabilities," and said it has patched infrastructure and updated evaluation controls. Hugging Face disclosed the intrusion on July 16 without naming the attacker, as reported by BleepingComputer in AIPD's July 21st edition; Tuesday's OpenAI post names the attacker for the first time.
Read at OpenAI ↗ · Read at LessWrong ↗
Treasury Secretary Scott Bessent will represent the U.S. in a Trump era bilateral AI dialogue with China this September, per Reuters. The meeting is expected before Chinese President Xi Jinping's planned Sept. 24 U.S. visit and was a stated outcome of the Trump-Xi summit in May. Dates, agenda, other participants and location have not been finalized.
Read at Reuters ↗
The Trump administration will overhaul roughly $200 billion in annual federal R&D funding to favor individual scientists and AI-driven research over universities, per drafts of a report from the White House Office of Science and Technology Policy (OSTP) and an OMB memo shared with the Wall Street Journal. Federal R&D agencies must submit implementation action plans within 90 days and fold the guidance into their FY2028 budget submissions. The report, titled "Science: A New Golden Age," anchors the pivot to national mission programs including the Genesis Mission for AI driven scientific discovery, commercial fusion power demonstration by the mid-2030s and next generation semiconductor technologies. A parallel OMB rule would give political appointees more power over grants to align research with administration priorities.
Read at WSJ ↗ · Read at The White House ↗
Treasury Secretary Scott Bessent said the administration would examine whether Chinese open-weight AI models have been distilled from American models and could sanction the firms responsible in a Fox Business interview Tuesday. "If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft," he said, and told the network that Treasury would examine the issue in the coming days or weeks. Bessent said Treasury has found watermarks of U.S. models on many Chinese models and raised the possibility of requiring companies deploying Chinese models to disclose that to customers. His remarks landed the same day Reuters reported he would lead the September U.S.-China AI dialogue. Anthropic wrote to the Senate Banking Committee last month alleging Alibaba had carried out "the largest known distillation attack" against it.
Read at CNBC ↗ · Read at Fox Business ↗ · Read at MeriTalk ↗
Anthropic will donate another $20 million to Public First Action, bringing its total midterm cycle commitment to $40 million, the company announced Wednesday. Public First Action, a 501(c)(4) advocacy group, is tied to three super PACs backing candidates across the political spectrum who support AI safeguards. Anthropic said the escalation was driven by the risks posed by increasingly powerful models, including its own Mythos, and said the money cannot be used to influence the election of any particular candidate. The spending sits opposite Leading the Future, the pro-industry super PAC network backed by OpenAI President Greg Brockman that ended June with $31 million on hand.
Read at WSJ ↗ · Read at Axios ↗
Sam Altman will brief the Trump administration and U.S. lawmakers next week on OpenAI's upcoming generation of models, OpenAI global public affairs head Chris Lehane said at a Washington briefing Tuesday. Lehane said the meetings will focus on capabilities relevant to work and scaling work. U.S. officials are finalizing a safety review framework expected in the next few weeks. The Trump administration is weighing whether to create an independent AI model regulator with industry input, a plan Treasury Secretary Scott Bessent helped develop that resembles a proposal outlined last week by Google DeepMind CEO Demis Hassabis. Lehane said OpenAI made significant changes at government request before releasing GPT-5.6 and is backing a Massachusetts frontier risk proposal as part of a "reverse federalism" push.
Read at Bloomberg ↗ · Read at Communications Today ↗