The evening briefing.

AI Capabilities and Risks. New research highlights the escalating capabilities of AI models, with a benchmark demonstrating that Claude Mythos and GPT-5.5 can autonomously develop real browser exploits. This development underscores significant security implications for web infrastructure and user data. Concurrently, OpenAI has introduced a new feature allowing ChatGPT to connect directly to users' bank accounts, providing access to detailed financial transactions. This expansion of access raises substantial privacy concerns and questions about the responsible handling of sensitive personal data by AI systems.

Economic Shifts and AI Costs. Data indicates that American jobs with significant AI exposure are beginning to disappear, signaling an early but potentially bleak trend in the labor market. The recent blockbuster IPO of AI chip darling Cerebras has intensified market hype for other major players like SpaceX and OpenAI, yet it simultaneously highlights how smaller innovators struggle for visibility and funding. The high operational costs of advanced AI are also evident, with one developer reportedly spending $1.3 million on OpenAI tokens in a single month. This disparity in resources contributes to a growing divide between the 'haves' and 'have-nots' in the AI gold rush.

Decentralized AI and Hardware Innovation. A US startup is pioneering a novel approach to distributed computing by offering homeowners payment to host small data centers on their property, aiming to decentralize AI infrastructure. In hardware advancements, a tiny new perception chip promises to enable skinny Aviator-style smart glasses with real-time spatial intelligence, eliminating the need for bulky GPUs. Further pushing on-device AI, researchers have developed electronic rings capable of interpreting multiple sign languages into text. Meanwhile, Japan is seeing a surge in demand for robot wolves, which are being deployed to deter record bear attacks.

AI Governance and Content Integrity. The scientific community is taking steps to address the misuse of AI in academic work, with the research repository ArXiv implementing a policy to ban authors for a year if they rely solely on AI for paper generation. In a move to combat misinformation and protect creators, YouTube is expanding its deepfake face-swap detection tool to all adult creators. This system allows users to request the removal of videos that use AI to falsely depict their likeness, reinforcing the platform's commitment to content authenticity.