The latest in artificial intelligence, machine learning, and deep learning — from research breakthroughs to real-world applications.
106 articles
Anthropic will give outside safety evaluators permanent employee-level access and the right to publish findings, as CEO Dario Amodei urges the industry to slow down AI development.
Anthropic's third AI misuse report discloses five biological case studies and cyber clusters linked to Russia, ShinyHunters and Chinese-speaking operators between December 2025 and August 2026.
A joint advisory from the NSA, FBI and CISA names DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun and Z.AI, while China's commerce ministry calls the distillation allegations baseless and threatens countermeasures.
Internal figures released two days after the essay show automated systems performing most of the daily labor inside the lab, an awkward backdrop for a public plea for collective restraint.
China's state-backed AI champion is pricing its new flagship aggressively and open-sourcing its small variants, betting that efficiency and national infrastructure will outweigh raw scale.
Bloomberg reported the cluster would be among the largest known Huawei installations, backing DeepSeek's models with domestic accelerators as US export curbs keep Nvidia's best chips out of China.
The model translated Andrew Wiles's landmark 1995 proof into more than 13 million lines of Lean code during an 11-day run, a scale that one mathematician said no human project had come close to finishing.
A reported first-of-its-kind dialogue between the two AI powers would lean on private labs to police their own models against misuse, two weeks before the leaders meet.
OpenAI is positioning the model as its most safety-hardened release yet, with real-time monitoring that escalates to human researchers within 30 minutes and a phased rollout that begins with enterprise cybersecurity customers before reaching consumer tiers.
An open-weight model above 300 billion parameters is now available to any developer who wants to self-host it, ten days after the same capabilities were limited to a paid API. The timing of the release, and the agentic benchmarks that accompany it, point to a strategic bet on software that acts rather than merely responds.
The release keeps base list prices unchanged while cutting prompt-cache reads from $1.00 to $0.25 per million tokens, and it pairs the widely available model with a less-guarded twin restricted to vetted US cybersecurity and life-science organizations.
OpenAI said late Friday it will stop supplying models to Cursor after SpaceX's $60 billion takeover, proposing a November 12 cutoff, while Anthropic moved to absorb displaced developers with more Claude compute and a 25 percent limit increase.
Google DeepMind's Co-Scientist evolved from a hypothesis generator into a closed-loop lab partner, synthesizing semiconductor thin films on its first attempt and cutting fabricated-results rates to 4%, according to a Nature-published study highlighted on August 28.
Two of the world's biggest music publishing houses filed suit against Anthropic, accusing the Claude maker of training on pirated songs and books and seeking up to $150,000 per infringed work. The legal fight could set the price every AI lab pays to train on music.
Anthropic agreed to a multi-year $45 billion compute deal with UK-based Nscale on August 26, the largest single-vendor commitment in the AI startup's history and the third multi-billion infrastructure agreement since June.
OpenAI disclosed measured results from its first-party Jalapeno ASIC on August 26, saying the 700W chip delivers 1.5-1.9x more AI work per watt and 1.7-3.6x lower latency than competing systems across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T.
Nvidia's $20 billion Groq acquisition became a shipping product on August 24, with the LPX rack now in volume production at neocloud Nebius and delivering 4x the inference throughput of the nearest competitor, Nvidia executives said.
Anthropic's Claude service went dark for thousands of developers worldwide on August 24, 2026, breaking Claude Code, the API and the chat product. The incident is the second major outage in three weeks for a frontier model API and revives questions about concentration risk in enterprise AI.
Anthropic has begun early investor outreach for an IPO that could raise more than $100 billion at a valuation as high as $2 trillion, the New York Times reported on August 23. Morgan Stanley, Goldman Sachs and JPMorgan are on the underwriting team, and a prospectus could land before the end of August.
Microsoft began rolling out a unified Copilot app for personal, work and school accounts on August 18 while revealing that its Cowork agent reached 58,000 active users inside Microsoft Digital within weeks of internal release. The two moves together redraw the company's consumer-and-enterprise AI strategy.
Internal documents seen by Bloomberg show Anthropic generated $11.5B in Q2 2026 versus OpenAI's $6.7B — a 14-fold year-over-year rise — driven by enterprise API revenue (80% to 85% of mix) and Claude Code's ~$8B contribution.