AI Trends: 16 October 2026

Safety: Anthropic warns about self-replicating models

Anthropic said small AI models could, within months, gain enough capability to copy themselves across internet-connected consumer devices such as smart TVs and appliances. Its tests on commercially available hardware found models can already perform many of the individual steps, but no model has yet completed the full process on its own. Separately, OpenAI and Anthropic both told a British parliamentary hearing on 13 October that they favor legally binding AI safety requirements, while Google DeepMind and Meta leaned toward existing regulation and technical standards. A Semperis survey adds a security angle: 84 percent of organizations hit by ransomware said attackers used AI somewhere in the attack, and 44 percent said criminals exploited their own AI systems.

Markets: rate fears hit AI stocks and CoreWeave

Hotter inflation revived expectations of tighter Federal Reserve policy, driving steep declines in tech and AI chip stocks, with losses spreading into Asia. CoreWeave was hit hardest: yields on some of its bonds approached 15 percent and its shares fell sharply. CEO Mike Intrator called the reaction irrational and pointed to the company's contract backlog. The Financial Times also reported that CoreWeave is exploring loans backed by computing equipment to attract life insurers. Against that backdrop, ASML posted record third-quarter revenue of 11.7 billion euros and forecast 2026 sales of 44 to 45 billion euros, with customers already discussing 2029 and 2030 equipment needs.

Companies and deals

The Financial Times reported Anthropic revenue of 16.8 billion dollars for the latest quarter, up 50 percent from the previous one, as the company prepares for a potential IPO. AI infrastructure company Modal raised a Series D that more than tripled its valuation from 4.65 billion to about 15 billion dollars. Nvidia committed 250 million dollars in compute over five years for AI safety and robotics research in the UK, and plans a minority investment in French healthtech firm Doctolib. DayOne's Malaysian data-center financing has grown to about 7.2 billion dollars, and Pluto Security raised about 42 million dollars to help companies control unauthorized employee-built AI agents.

Legal and legal-adjacent

Apple is suing OpenAI over alleged trade-secret misuse. OpenAI denies the claims and has voluntarily agreed not to use the disputed information while the case proceeds. Meanwhile the FBI has charged an OpenAI engineer with hiding an audio recording device in a San Francisco executive conference room; the allegations are unproven.

Products and tools

Google's Gemini 4 Argon is now broadly available to developers, businesses and subscribers, with promotional API pricing of 2 dollars per million input tokens and 10 dollars per million output tokens. xAI released Grok Voice 3.0, a speech-to-speech model with sub-second latency and unchanged API pricing. Apple announced the 299 dollar HomeView smart-home hub with Siri AI, shipping 13 November but not initially in the EU. OpenCode 2.0 adds background subagents, and Nvidia introduced cuPotential, a CUDA-X library for agents that simulate atomic and molecular interactions.

Society

CBS reported on retailers using AI and personal data for individualized pricing, and the BBC found candidates applying to more than 200 jobs with AI, pushing employers to screen harder. Two physicians argued that clinicians must stay accountable for treatment decisions even as AI diagnosis improves.

Compiled from public reporting for the 48 hours ending 16 October 2026.