Here is today's AI news. Story 1. OpenAI. Sam Altman reportedly told employees that OpenAI is open to slowing AI development amid growing concerns about the safety of advanced systems. (September 11). **OpenAI Considers Slowing Frontier AI Development** Sam Altman reportedly told employees at an all-hands meeting that OpenAI may slow development of its most advanced AI systems, ideally coordinating with rival labs, though some may refuse. The discussion reflects rising concern over systems capable of autonomous cyberattacks and other catastrophic risks. Altman said further capability advances should wait for progress in monitorability, alignment and understanding model behavior. Story 2. Meta. Meta released Muse, a consumer-focused AI agent designed to handle tasks such as booking travel, shopping, managing email, and completing online forms. (September 11). Meta unveiled Muse, a consumer AI agent that can handle tasks like booking travel, shopping, sending emails and filling out forms. The company says it runs on a secure virtual machine, connects to apps like email, calendar and payments, and can use Stripe’s Link for checkout. It matters because Meta is pushing AI from chat into action, competing in the emerging agent race. Story 3. Anthropic, Google DeepMind and OpenAI. The three AI companies began cooperating on AI-safety efforts, including work on evaluating and mitigating risks from increasingly capable models. (September 15). On September 15, OpenAI confirmed it had held several weeks of AI-safety talks with Anthropic and Google DeepMind. The discussions may support shared safety standards, independent model evaluations and risk-mitigation practices for increasingly capable systems. OpenAI policy chief Chris Lehane said, “It’s better to try to work together to prioritize safety,” adding that an antitrust waiver was unnecessary. The cooperation could improve consistency across frontier labs, though no formal agreement was announced. Story 4. OpenAI and Microsoft. A U.S. appeals court upheld a ruling favoring OpenAI and Microsoft in part of a lawsuit alleging that GitHub-hosted code was misused to train generative AI systems. (September 16). On Sept. 16, the Ninth U.S. Circuit Court of Appeals upheld dismissal of DMCA claims against Microsoft, GitHub and OpenAI in *Doe v. GitHub*. Developers alleged Copilot and Codex trained on GitHub code and produced snippets without attribution. The court said the tools generate “new works,” rather than removing copyright information from existing ones. Contract claims remain pending, a ruling that could influence other AI-training lawsuits. Story 5. Anthropic. Anthropic published a major threat-intelligence report detailing attempts to misuse Claude in cyberattacks, influence operations, surveillance. Anthropic’s September 2026 threat-intelligence report details attempts to misuse Claude between December 2025 and August 2026 for cyberattacks, influence campaigns, surveillance, fraud, biological research and weapons development. Russian-linked hackers reportedly automated much of an attack chain against more than 20 organizations, while Iranian actors built surveillance tools, including a malicious Firefox extension. Anthropic says it disrupted every operation and shared findings with authorities and AI companies, underscoring the need for stronger safeguards as AI shifts from advising attackers to executing operations. That's all for today's AI news briefing.