AI news, 11 September: Anthropic disrupts bioweapons and state misuse of Claude

Anthropic details blocked misuse of Claude for bioweapons research and state operations. In its September threat-intelligence report covering late 2025 to mid-2026, Anthropic said it disrupted multiple cases in which users tried to use its Claude models for research that could support biological weapons development, including gain-of-function work on viruses and toxin redesign, plus cyber operations, surveillance, influence campaigns and conventional weapons software. The company highlighted five biological case studies, stressed it could not confirm harmful intent in every instance, and noted it shared findings with authorities while strengthening safeguards. Coverage from the Associated Press, BBC, CNN, NPR and The New York Times underscored that biological misuse ranks among the most serious risks of frontier AI, even as the same capabilities aid legitimate science. Anthropic also reported Chinese and Russia-linked operators circumventing restrictions.

California enacts first-in-nation AI auditor framework. Governor Gavin Newsom signed SB 813 and AB 1405, creating a framework for independent verification organizations to assess AI systems and a state registry of AI auditors with rules on independence, transparency and ethics. The measures establish a California AI Standards and Safety Commission for voluntary standards. Anthropic had backed the package earlier and OpenAI endorsed it shortly before signing; Newsom called for matching federal rules. Separate child-safety chatbot measures were also signed around the same time.

OpenAI launches Agents API in public beta and ChatGPT for finance. OpenAI released a managed Agents API that exposes the Codex harness for orchestration, long-running sessions, context management and tools, letting developers build agents without handling the full infrastructure. It supports OpenAI-hosted or partner sandboxes with no extra API fee beyond tokens and tools. Separately, ChatGPT for Financial Services launched on GPT-6 Astra with built-in market data connectors and design partners including Morgan Stanley and Evercore, targeting tasks such as pitchbooks and earnings analysis.

US agencies warn of industrial-scale Chinese AI distillation. The NSA, CISA and FBI issued a joint advisory alleging that Chinese firms including DeepSeek, Moonshot AI and Alibaba have systematically extracted capabilities from Western models such as Claude, GPT, Gemini and Grok via large-scale distillation since at least late 2024, likely with government awareness. The advisory is guidance rather than a ban and urges detection, response modification and information sharing.

Other notable moves. Apple’s first event under new CEO John Ternus emphasized the iPhone as an AI device and unveiled a foldable starting at $1,999. Meta continued rollout of its Muse personal agent. Funding activity included sizable raises in the AI chip and legal-AI sectors.