AI Briefing

Oregon just passed the country's toughest AI chatbot law complete with a private right of action, Eli Lilly is betting $2.75 billion that AI-designed drugs are ready for primetime, and Apple is about to let you swap ChatGPT for Claude or Gemini inside Siri. Plus, an outside team spent three weeks trying to break Anthropic's agent monitoring and found some cracks worth knowing about.

01PolicyPYMNTS ↗

Oregon Passes Nation's Toughest AI Chatbot Law With Private Right of Action

Oregon legislators gave final approval to SB 1546, a bill requiring AI chatbot operators to implement safety protocols, disclose that users are interacting with AI, and detect signals of suicidal ideation or self-harm to interrupt conversations and direct users to crisis resources. The bill includes stricter rules for minors and, critically, allows individuals to sue operators for $1,000 per violation. It passed the Senate 26-1 and the House 52-0 with near-unanimous bipartisan support. This is the first US chatbot-specific law with real enforcement teeth, and with similar bills advancing in Washington and Utah, it could set the template for how states regulate AI interactions nationwide.

02IndustryCNBC ↗

Eli Lilly Signs $2.75 Billion AI Drug Deal With Insilico Medicine

Eli Lilly signed a deal worth up to $2.75 billion with Hong Kong-based Insilico Medicine to bring AI-designed drugs to the global market, with $115 million upfront and the rest tied to regulatory and commercial milestones. Insilico has developed 28 drug candidates using its generative AI platform Pharma.AI, with nearly half already in clinical trials. Lilly gets exclusive worldwide rights to develop and commercialize novel oral therapeutics across several disease areas. This is the largest AI drug discovery deal to date and a strong signal that pharma giants now view AI-designed molecules as commercially viable, not just research curiosities.

03Product9to5Mac ↗

Apple to Open Siri to Claude, Gemini, and Other AI Chatbots via Extensions in iOS 27

Apple is developing a new Extensions system for iOS 27 that will let any AI chatbot app from the App Store integrate directly with Siri, ending OpenAI's exclusive partnership. Users will be able to choose which AI model handles Siri queries through a dedicated Extensions menu in settings. The change, expected to be announced at WWDC on June 8, means Claude, Gemini, and potentially any third-party AI could run inside Siri across iPhone, iPad, and Mac. This is a massive distribution play that could reshape the competitive landscape for AI assistants, turning Siri from a single-vendor product into an AI marketplace.

04ResearchMETR ↗

METR Red-Teams Anthropic's Agent Monitoring Systems, Finds Vulnerabilities

AI safety organization METR spent three weeks adversarially testing Anthropic's internal agent monitoring and security systems, many of which were described in the Opus 4.6 Sabotage Risk Report. The exercise discovered several novel vulnerabilities, some of which Anthropic has since patched. Importantly, METR noted that none of the findings severely undermine the major claims in the Sabotage Risk Report. This is one of the first public examples of an independent organization conducting adversarial red-teaming of an AI company's monitoring infrastructure, setting a precedent for the kind of third-party security audits the industry will need as AI agents become more autonomous.