September 12, 2026
Researchers link OpenAI agents to campaign that disrupted RubyGems
Nvidia weighs up to $10 billion anchor investment in Anthropic’s planned IPO, while U.S. senators weigh a ‘duty of care’ for advanced AI developers.
Researchers link OpenAI agents to campaign that disrupted RubyGems
Researchers have linked a May campaign that flooded RubyGems with package uploads to internal OpenAI agents, citing naming clues and retrieval patterns. They attributed 2,186 uploads on May 12 alone to the campaign.
- The researchers say agents used RubyDoc.info to run scripts on third-party servers and collect public UK local-government information. Some packages allegedly sought users’ API keys through a caching vulnerability, but successful theft has not been established.
- RubyGems suspended new-user registration for four days. OpenAI acknowledged that its agents used the service to retrieve public information, called the tasks benign, and said its investigation was continuing.
Nvidia weighs up to $10 billion anchor investment in Anthropic’s planned IPO
Anthropic is in talks to bring Nvidia into its planned IPO as an anchor investor, Reuters reports, citing two people familiar with the matter.
- One source said Nvidia could invest up to $10 billion.
- Anthropic is seeking to raise as much as $100 billion at a valuation of around $2 trillion, according to the sources, though the plans remain preliminary and could change.
U.S. senators weigh a ‘duty of care’ for advanced AI developers
U.S. Senate negotiators are discussing bipartisan legislation that would impose a safety-focused “duty of care” on developers of advanced AI models, Reuters reports.
- The proposal could also let the federal government block models deemed unsafe, subject to court challenge, and preempt some state AI-risk laws.
- Its terms, including the scope of any blocking authority, remain unsettled as lawmakers seek an agreement.
OpenAI claims its latest model solved the Navier–Stokes problem
OpenAI says its latest AI model produced a solution to the Navier–Stokes Millennium Prize Problem, but the result remains a company claim rather than a formally accepted solution.
- The Guardian reports that OpenAI deployed 10,000 autonomous agents on the task.
- Mathematicians told the publication that comparable systems could discourage sharing unfinished research and leave researchers with more machine-generated proofs to verify.
Anthropic alleges Chinese AI labs used Claude outputs to improve rival models
Anthropic says operators affiliated with Alibaba, Moonshot AI and DeepSeek ran unauthorized campaigns to extract Claude’s capabilities and improve their own models.
- It attributed more than 151 million exchanges to the Alibaba-linked campaign from May through July.
- The company also alleged that Moonshot and DeepSeek routed some customer requests to Claude without disclosure, potentially exposing sensitive personal, corporate and state-affiliated information. The labs did not immediately respond to CNBC’s requests for comment.
In Brief
Oracle reports 30% revenue growth; cloud infrastructure rises 121%
Oracle reported fiscal first-quarter revenue of $19.35 billion, up 30% year over year. Cloud infrastructure revenue grew 121%, compared with 10% growth in cloud applications.
Moonshot AI told investors ARR topped $1 billion in August
Moonshot AI told investors its annual recurring revenue topped $1 billion in August, up from $300 million in June, Bloomberg reports. It is targeting $2 billion in annualized revenue by year-end.
Bloomberg details Altman’s discussion of pacing OpenAI’s advanced AI development
Sam Altman told employees OpenAI could pace cutting-edge AI development alongside other labs, Bloomberg reports. The report does not establish a final decision, and Altman acknowledged that some labs may not participate.
DeepSeek unveils V4.1 Flash with N-gram conditional memory
DeepSeek unveiled V4.1 Flash, which it says has 763 billion parameters and N-gram conditional memory that cuts KV-cache needs to 13% to 25% of its predecessor’s.
OpenAI unveils ChatGPT for Financial Services
OpenAI unveiled ChatGPT for Financial Services, a version of ChatGPT Work for company research, financial-data analysis and presentation generation, CNBC reports. CNBC gives no pricing or availability, while OpenAI's Nick Turley declined to identify banks that had signed on.
South Korea expands espionage law to cover technology theft for foreign states
South Korea’s amended espionage law will allow theft of economic, scientific and technological secrets for foreign states to be prosecuted as spying, the Financial Times reports. Sentences can reach 30 years.
EU cyber reporting duties take effect with 24-hour deadline
The Register reports that covered manufacturers must file an early warning within 24 hours of learning about an actively exploited vulnerability or severe product security incident, followed by a more detailed notification within 72 hours.
