Breaking news and updates from the world of technology.
Kimi K3.1 Pro and a faster Highspeed variant bring a 1-million-token context window and sharply lower hallucination rates, with full open weights due by November 4.
OpenAI announced it will permanently replace live internet access during training with cached web copies and simulated environments, following months of incidents including a September sandbox escape and the July Hugging Face breach.
OpenAI and Google announced a deal putting GPT-6 Astra and the Codex coding agent in Google Cloud's Model Garden, five months after OpenAI's exclusivity arrangement with Microsoft expired.
The funding round would be the largest private company valuation in history, nearly doubling OpenAI's March 2026 price tag of $852 billion post-money.
Anthropic’s new Claude Duet app works in the background across iOS, Android, and the web, handling tasks and surfacing what needs your attention while you’re busy.
Anthropic will give 5% of its equity, growing automatically with its valuation, to a new fund aimed at AI's societal costs — citing job displacement concerns and what CEO Dario Amodei calls a crisis of public trust.
Anthropic’s AI model solved a 26-year cryptographic mystery, tracing NIST’s elliptic curve parameters to a Unix timestamp-seeded C random number generator from 1997 — making deliberate backdoors far less plausible.
India's cybersecurity agency says AI agents from several providers reached confidential business records and personal data at Indian organizations while running evaluations, exceeding what they were authorized to do.
Palo Alto startup Elorian released its first model, Glimpse, which returns to crop and re-inspect source images mid-task instead of converting them to text once and reasoning from there.
Meta has open-released its most capable AI model yet — 816 billion parameters, multimodal, and designed to run locally on Apple silicon, though you will need at least 1TB of RAM.
An open letter signed by roughly 400 scientists from 34 countries calls for an international body with power to set AI containment standards and maintain a public registry of serious security incidents.
Anthropic's red team ran open-weight models across 15 budget Android TV boxes and simulated network environments, finding that autonomous self-replication across IoT devices is months away, not years.