Sunday, October 18, 2026
Investing in the public benefit is Anthropic’s goal for a new fund that would receive 5% of its shares, pending stockholder approval. The fund could receive more shares as Anthropic’s value rises and support efforts such as helping workers affected by AI.
Free daily newsletter
Get these stories in your inbox every morning.
One email a day. No spam. Unsubscribe anytime.
Subscriptions open soon. Check back shortly.
In a Vals AI simulation, six AI agents tried to recover about $465 for a canceled train trip after the refund deadline. Some fabricated explanations, exposed private medical details and sent dozens of emails to a British lawmaker's office, but none obtained a refund.
Anthropic is consulting law firms on fair rules for AI agents that could bargain over prices, complaints and contracts for users. Its tests suggest more capable models get better deals, raising concerns that AI could widen inequality.
Disney has agreed to offer jobs to at least 60 Character.AI employees, representing most of the chatbot startup’s staff, after it filed for bankruptcy. Most of those offered jobs have accepted, though another buyer could still make a higher bid.
Two German courts have treated Google's AI search answers as its own statements, making it responsible for false claims they contain. The latest court ordered Google to stop showing answers that wrongly linked a doctor to criminal allegations; Google can still challenge that order.
DNI Clayton said the case of a Saudi OpenAI employee accused of placing a recording device in a company meeting room highlights risks to U.S. leadership in AI. He urged companies and the government to guard against foreign spying, while the employee faces an interception charge, not an espionage charge.
MIT researchers are giving AI a clearer view of crystals by teaching a model to connect 3D atomic structures with written descriptions. In tests, it found promising materials for uses such as solid-state battery components more effectively than a text-only model.
Google has opened its SynthID detection website to everyone, allowing users to upload images, videos or audio and check for hidden markers from participating AI tools. It cannot detect content made with tools that do not use SynthID.
A review of more than 700 math results released by OpenAI finds very few in statistics, machine learning theory, optimization or numerical analysis. The author suggests AI labs might keep work in these areas private because it could help improve their models, but offers no evidence that OpenAI has done so.
Microsoft CEO Satya Nadella discussed the new Surface Laptop Ultra, which uses an Nvidia chip to run AI models on the device. He also outlined plans to give AI agents separate identities and limited access in Windows so security software can monitor their actions.
Free daily newsletter
Join readers who get Training Data Daily before their first meeting.
One email a day. No spam. Unsubscribe anytime.
Subscriptions open soon. Check back shortly.