AI News Roundup – OpenAI discloses more “rogue AI agent” incidents amid calls for accountability, France’s top literary prize drops top contender over AI use allegations, Chinese companies developing alternative to Nvidia software for AI chips, and more

Practices & Technologies

Artificial Intelligence

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

  • OpenAI has disclosed several further incidents of “rogue” AI agents, sparking further calls for regulation and accountability for AI companies whose models act maliciously, according to Reuters. In recent weeks, OpenAI disclosed further incidents where its models, acting autonomously, had accessed systems without authorization, including an Australian government health data portal this summer, sparking an outcry in that country over the company’s delay in informing authorities about the breach. The company also disclosed incidents where models accessed U.S. government data, but stated that there was no indication that such access was unauthorized. OpenAI has acknowledged the need for additional transparency, though sources told Reuters that the company’s internal investigation into AI behavior was locked down and shaped by the company’s attorneys. Separately, The Wall Street Journal reported this past week that OpenAI had fired three researchers who had shared confidential AI safety information with third-party safety researchers. Since late July, when the first “rogue agent” incidents were disclosed, worries have risen in the AI industry over the ability to control model behavior and provide AI safety. As this AI Roundup has reported in recent weeks, leading AI labs called for a slowdown in AI development to address safety concerns. At a recent White House event, U.S. President Donald Trump promoted a self-policing AI safety accord, signed by the president along with Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang, and SpaceX leader Elon Musk. The accord states that the companies agree to implement “robust internal controls,” partner with an “independent external auditor” to assess whether those internal controls were being followed, and establish a board-level committee at each company to evaluate reports from internal and external auditors. Meanwhile, other U.S. government bodies are moving for more concerted action on AI regulation. The Federal Trade Commission has opened an investigation into Anthropic and OpenAI over whether they have lied to consumers about AI harms, while Senators Josh Hawley, a Republican of Missouri, and Chris Murphy, a Democrat of Connecticut, will soon introduce legislation that would hold AI companies criminally and civilly liable for hacking incidents.
  • Le Monde reports that France’s top literary prize has dropped a leading contender from consideration after AI use and plagiarism allegations surfaced against the author. The organizers of the Prix Goncourt announced this past week that the bestselling and critically acclaimed book “C’était Ça ou Mourir” (“It Was That or Dying”) by the Haitian- Québécois writer Thélyson Orélien would be removed from contention for the award after allegations arose on the social media network X that Orélien had used AI in writing the novel, though different AI detectors (including Pangram, which this AI Roundup covered last month) provide different results. Orélien denied using AI, but the prize stood by its decision, claiming that it was “driven by the desire to preserve the integrity of the Prix Goncourt and the Academy itself, whose central role is to promote and honor literature written by women and men.” Reporting by La Presse and Radio-Canada in Quebec, where Orélien resides, also found evidence that the author plagiarized portions of a 2013 article and a 2010 novella he authored. Orélien has claimed that the campaign against him is an “attempt to silence my voice” and that it was racially motivated, and that he is working with his publishers to compile evidence that he wrote the book himself.
  • CNBC reports on Google DeepMind’s announcement of Gemini 4 Argon, the company’s newest and most advanced AI model. According to a company blog post, the new model beats several other leading models, including OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 in a variety of benchmarks, including those measuring knowledge work ability and coding performance. The company said that Gemini 4 Argon is already used internally to optimize memory usage at Google datacenters, migrate C and C++ codebases to the Rust programming language, and assist in the company’s quantum computing research. A Google executive told CNBC that the model was “incredibly well-rounded,” and that its cybersecurity features were greatly improved over predecessor models, saying that “[w]e really believe that a model of this caliber and this level of frontier performance is meaningfully important for defenders.” The company said that Gemini 4 Argon will be rolled out in phases, beginning with partnered cybersecurity researchers in the company’s Fairwind Program, and the company also announced its intention to engage in the U.S. government’s voluntary model safety testing process (a program that this AI Roundup covered earlier this year). Feedback from this testing will help improve the model before it is released to developers, enterprise customers, and consumers, as a public release date was not announced.
  • A U.S. appeals court has ruled against Anthropic in its challenge to the Pentagon’s designation of the company as a supply chain risk, according to Bloomberg. The decision by a split three-judge panel of the United States Court of Appeals for the District of Columbia Circuit held that the U.S. Defense Department acted within its authority to bar the AI lab from Pentagon contracts. Anthropic’s dispute with the U.S. military burst into public view in February of this year, when the company raised concerns over the potential for its Claude AI technology to be used to surveil American citizens or for weapons targeting without enough human oversight, as this AI Roundup covered at the time. The dispute led to the Pentagon designating the company as a supply chain risk, essentially banning the company from Defense Department contracts. Anthropic challenged this decision in court, saying that the Pentagon had not adequately explained the decision as required by the Supply Chain Security Act. Rejecting these arguments, the D.C. Circuit ruled that the Pentagon had “reasonably concluded that removing Anthropic from the Department’s supply chain was necessary to protect national security by reducing supply chain risk to the Department’s information systems.” In dissent, Judge Karen LeCraft Henderson argued that the court had read the governing law too narrowly and that Congress intended for the law to combat malicious actors infiltrating government supply chains. In a statement, Anthropic said that the company “remain[s] confident in our position and are considering all options, including further review,” noting that an order from a federal judge in San Francisco in a related lawsuit that lifted the Pentagon’s ban remained in place for now. Anthropic CEO Dario Amodei recently met with President Trump at the White House, with the latter stating that he “liked [Amodei] and his wife a lot,” possibly indicating a thawing of relations between the federal government and Anthropic as the litigation continues.

Patent Drafting Using Generative AI: Strategic and Doctrinal Considerations

MBHB Partner Manav Das on How Apple’s Lawsuit Against OpenAI May Shape Future AI Litigation in Law.com

AI News Roundup - Chinese President Xi Jinping and President Trump meet at summit with AI on the agenda, Oracle data center megaproject flounders in New Mexico, AI labs sued over alleged antitrust violations, and more