Here is today's AI news. Story 1. Anthropic. CEO Dario Amodei urged AI companies to slow the pace of frontier-model development, proposing independent safety evaluators, industry coordination, and international cooperation. (Sept. 12). Anthropic CEO Dario Amodei urged AI firms to “slow the pace” of frontier-model development in a Sept. 12 essay, arguing safety efforts need time to catch up. He proposed independent evaluators with employee-like access, industry-wide standards, and international coordination. “Progress will still seem fast,” he wrote, but pacing could reduce the risk of “something goes seriously wrong.” Story 2. OpenAI. CEO Sam Altman said he agreed with Amodei’s call for “pacing” AI progress, while clarifying that it should not mean stopping development. (Sept. 12–14). OpenAI CEO Sam Altman backed Anthropic CEO Dario Amodei’s call to “pace the frontier” of advanced AI, not halt it. Altman said OpenAI would adopt independent evaluators with employee-like access to models, tools and researchers. The move signals growing industry support for external safety testing as frontier-model capabilities accelerate. Story 3. OpenAI. The company reportedly launched Astra for Law, a GPT-6 Astra-based legal-AI platform featuring law-firm workflows and client controls. (Sept. 18). **OpenAI launches Astra for Law:** OpenAI introduced a legal configuration of GPT‑6 Astra for law firms and legal-tech vendors. It combines legal-analysis instructions with a U.S. legal search index spanning 230M+ URLs. Initially available to selected firms via Trusted Access in ChatGPT and Codex, it adds confidential-work controls; API access is planned. OpenAI says it scored 54% on Vals AI’s legal-research benchmark, versus 38.7% for Astra using web search alone. Story 4. Google. Google expanded its Gemini 3.8 Live lineup with live-dialogue and extended-thinking models supporting real-time voice interactions and broad multilingual capability. (Sept. 15). Google launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on Sept. 15, expanding its voice AI lineup. The models power real-time, natural dialogue; the extended-thinking version handles complex, multi-step tasks while speaking. Google says Live can detect and switch between 97 languages mid-conversation, and both models support background tool calls and visual grounding. Story 5. Meta. CEO Mark Zuckerberg rejected a coordinated industry-wide AI slowdown, arguing that AI safety should remain the responsibility of individual companies rather than a collective development pause. (Sept.. On Sept. 15, 2026, Meta CEO Mark Zuckerberg rejected calls from Anthropic, OpenAI and others for a coordinated AI-development slowdown. He argued competition, user trust and legal liability already incentivize labs to build safely: “Any lab that doesn’t focus on alignment will fall behind.” Zuckerberg cited Meta’s months-long delay of its Muse AI agent for safety and security work, underscoring a market-led, company-by-company approach. That's all for today's AI news briefing.