📢 200+ AI experts urge governments to act now 📢
Over 200 experts call for urgent regulation before AI reshapes the global labor market.

Over 200 AI researchers and economists, including 16 Nobel laureates, signed "We Must Act Now", a Stanford-organized statement urging governments to build safety nets and labor policy for an economic shift the signatories compare to the Industrial Revolution. The statement argues AI will grow "radically more powerful" within 10 years, with signatories including Google's Jeff Dean, Anthropic's Jack Clark, and OpenAI's Noam Brown. Economist Anton Korinek warned that unlike past technology shifts, AI may leave only a few years to adapt.

🤖 Hassabis pitches U.S. safety body for frontier AI 🤖
Google DeepMind's CEO calls for regulation before frontier AI rollout spirals.

Demis Hassabis published a proposal for an independent, U.S.-based body — modeled on FINRA — that would safety-test frontier AI models before public release, with labs submitting models 30 days before launch for review of risks like deception, bioweapons creation, and hacking. Hassabis warned open-source capabilities could reach dangerous territory within 18 months and urged the body to be operational this year, including provisions to coordinate a slowdown among frontier labs if reviewers deem it necessary.

🚀 Moonshot ships Kimi K3 with 2.8T parameters 🚀
China's frontier model matches Western pricing and agentic speed.

Moonshot AI shipped Kimi K3, a 2.8-trillion-parameter multimodal model with a 1-million-token context window, delivering benchmark performance competitive with Claude Fable 5 and GPT-5.6 Sol while matching Claude 5 Sonnet's pricing at $3/$15 per million tokens. In a striking demo, K3 worked autonomously for 48 hours to design and verify a chip capable of running a miniature version of itself. Open weights are scheduled for release on July 27.

INTERESTING TO KNOW

🤖 Anthropic extends Claude Fable 5 access through July 19 🤖

Anthropic extended access to Claude Fable 5 and raised Claude Code usage limits through July 19, as OpenAI simultaneously removed usage caps on GPT-5.6 Sol — a sign the frontier model wars are increasingly fought on availability and pricing rather than raw capability alone.

🤖 OpenAI ships GPT-Red to stress-test GPT-5.6 Sol 🤖

OpenAI published research on GPT-Red, a model trained to generate adversarial prompts and uncover vulnerabilities in other AI systems at scale. Feeding its attacks back into training cut failures on a prompt-injection benchmark by 6x for GPT-5.6 Sol, pointing to automated red-teaming as a new industry standard for safety testing.