Bayer won a unanimous St. Louis verdict in its first Roundup cancer trial since a Supreme Court ruling narrowed plaintiffs’ options. The decision may shape 65,000 claims as Bayer pursues a proposed $7.25 billion settlement.

OpenAI has updated its Model Spec, defining five values for model behavior and a hierarchy for resolving conflicting instructions. The guidance sets firm safety limits, protects human control and privacy, and acknowledges today’s models don’t yet fully meet its standards.

Michigan’s Senate race pits Democrat Abdul El-Sayed’s affordability agenda against Republican Mike Rogers, as attacks on El-Sayed’s loyalty and views sharpen debate over identity and pluralism. A poll shows El-Sayed leading 49% to 42%.

Anthropic added a silent, after-the-fact safety check that flags risky Claude chats for human review, complementing safeguards that block harmful requests in real time. It catches most high-priority cases, but attackers can still evade it.

Thrive Holdings will buy ADT for $10 a share, taking the home-security giant private in a $15.3 billion deal. It plans to use AI to improve alerts, reduce false alarms and spot equipment problems sooner.

Amazon has added more than 100 Alexa shopping tools, letting U.S. customers manage lists, subscriptions, deliveries and account issues from its app or website. The assistant can also track prices and buy items when they hit a target.

OpenEvidence’s clinical AI is now embedded in NYU Langone’s Epic system, letting clinicians ask natural-language questions and get answers grounded in medical research and patient records. The rollout spans hospitals, clinics and medical trainees.

Microsoft is positioning its Execution Containers as a security boundary for AI agents, separating their intelligence from system access. The pitch comes as risks mount; Microsoft also beat quarterly estimates and forecast stronger sales.

Apple’s eight-week hardware push spanned phones, Macs, tablets and its first foldable iPhone, with major display upgrades. Eight products were in stores by late October; several new Macs and an iPad were due November 6.

Anthropic says an unreleased Claude model built five exploit chains across Chrome, Safari, Android, Windows and Linux, showing how flaws can combine to breach security boundaries. The tests were controlled, issues were addressed, and the model’s exact identity and level of human help remain unclear.