27 articles in this category
Anthropic says it disrupted Claude misuse tied to bioweapons research, Russia-linked cyber espionage against Ukraine, and attempts to extract Claude’s capabilities. A missed hacking incident and a call from Dario Amodei are pushing the question of model-development speed into the open.
Apple's Siri AI beta brings context from email, messages, calendar, photos, and notes to iOS 27—with onscreen awareness, cross-app actions, and a local-first privacy model.
Anthropic released Fable 5.1 and Mythos 5.1 as the same model with different safeguard levels, while lower cache-read pricing changes the economics of long-running agents.
Google’s third Flash-tier release in six weeks brings software engineering, agentic tasks, multi-step reasoning, and a 1-million-token context. Gemini 3.8 Flash Cyber adds vulnerability discovery and patch generation behind the Fairwind Program, while its introductory API price doubles in January.
Google rolled out Gemini 3.8 Live with cost-efficient real-time voice and visual grounding, plus an Extended Thinking mode that reasons while it speaks. With Gemini integrations expanding and OpenAI shipping a competing full-duplex API, voice agents are moving toward the default interface for hands-busy moments.
OpenAI is phasing in GPT-6 Astra after it became the first model to trigger the company’s highest internal cybersecurity threshold — a preview of staged access as the new normal.
Meta's US-only Muse can act across email, travel, purchases, forms, and long-term goals, making personal knowledge and permission design the real product.
Mistral AI raised €3 billion at a post-money valuation above €21 billion, the largest equity round in European tech, to fund independent European compute, frontier research, and commercial scale.
A joint US advisory alleges six China-based AI companies are distilling US frontier models at industrial scale. China calls the accusations groundless; no court has ruled on them.
An internal OpenAI system produced an analytical proof that smooth solutions to Navier–Stokes develop a finite-time singularity, plus a machine-checked Lean formalization. The math community is excited and skeptical at once.
Crunchbase's half-year report shows global VC at an all-time high, with OpenAI and Anthropic alone absorbing $217 billion and AI claiming roughly two-thirds of all deployment.
Anthropic's new flagship product wires 60+ scientific databases and computation tools into an autonomous research agent — and Anthropic is using it to hunt drugs for neglected diseases.
Anthropic's Sonnet 5 launches at $2 per million input tokens, approaches Opus 4.8 on agentic benchmarks, and becomes the default for Free and Pro users.
The Council gave final approval to the AI Act omnibus on June 29 — extending SME relief to 750-employee companies, shortening transparency grace periods, and banning deepfake nudity tools in December.
The US lifted export controls on Claude Fable 5 and Mythos 5 on June 30. The restoration comes with a jailbreak severity framework, a bug bounty, and a lesson about AI as infrastructure.
OpenAI's GPT-5.6 family previews with trusted partners: Sol sets a Terminal-Bench record, Terra undercuts on price, and Luna chases speed — with 750 tokens per second on Cerebras.
Elon Musk confirmed Grok 4.5 is running inside SpaceX and Tesla on a new 1.5-trillion-parameter V9 foundation — with V9-based releases planned monthly through 2026.
Meituan open-sourced a 1.6T MoE with native 1M context under MIT license, trained entirely on a 50,000-card domestic cluster — after it secretly topped OpenRouter as 'Owl Alpha'.
Meta announced a cloud unit selling spare GPU capacity and hosted Llama APIs, expanded its NVIDIA deal to multi-gigawatt scale, and sent AI cloud stocks tumbling.
OpenAI proposed giving the US government a ~$42.6 billion equity stake via a sovereign wealth fund, and envisions Anthropic, Google, and Meta doing the same.
Enterprise AI agents need runtimes for state, tools, approvals, lineage, retries, and governance.
The Fable 5 dispute shows that frontier model safety now includes export controls, red teams, and operational shutdowns.
Apple's Gemini-backed Siri push shows why personal AI assistants are becoming operating-system infrastructure.
New cyber-range benchmarks show why AI security evaluation needs realistic multi-host attack and defense workflows.
Gemini Live API gives developers a stateful WebSocket path for voice, video, tools, and low-latency AI agents.
Gemma 4's open models, larger context, and device-friendly variants make private local AI workflows easier to ship.
Microsoft's MAI-Thinking-1 shows why owned reasoning models matter for Copilot, Azure, and developer workflows.