Archive
Latest — page 3
Older stories, newest first.
Anthropic says attackers used Claude to run whole cyberattacks
A September threat report details Russia- and China-linked campaigns and accuses seven Chinese labs of harvesting Claude's answers.
OpenAI launches Agents API in public beta
The API gives developers the Codex harness behind one call, with no extra fee beyond model, tool and container costs.
Apple unveils A20 Pro, first 2-nanometre smartphone chip
The chip debuts in the iPhone 18 Pro line and Apple's first foldable, the iPhone Duo, with double the AI compute of its predecessor.
OpenAI's Navier–Stokes proof: what 10,000 AI agents actually changed
The mathematics is one page in a longer ledger. The bigger story is a 10,000-agent swarm, a Lean-checked proof, and a cost line that keeps falling.
OpenAI scientist says no lab has solved AI alignment enough to race ahead
Jakub Pachocki wrote that continued maximum-speed scaling is not yet responsible and called for voluntary slowdowns until shared safety bars exist.
Four frontier releases in seven days: the cadence is now the strategy
Anthropic, Meta, Google and OpenAI all shipped in the same week. The buyers doing the evaluating are the ones absorbing the cost.
DeepSeek plans 160,000 Huawei Ascend chips for Inner Mongolia site
Bloomberg reported the order would create one of the largest known clusters of Huawei AI chips, used to run models rather than train them.
Sanders and Casar unveil bill to ban superintelligent AI
The proposed Ban Artificial Superintelligence Act would pause frontier AI development and impose 20-year prison terms on violators.
GPT-6 Astra saturates the benchmarks — and the alignment number is the one to read
98% on FrontierMath Tier 4, 100% on ExploitBench, a million-token context and premium pricing. The most consequential figure in the launch is 48% versus 0%.
Nvidia buys the open-model commons: inside the $12.93bn Hugging Face deal
The company that sells the shovels now owns the map. What changes when the dominant accelerator vendor controls where open models are found, compared and deployed.
Google launches Gemini 3.8 Flash Cyber under vetted Fairwind Program
The security-tuned model pairs with Google's CodeMender harness but is restricted to approved government and industry partners.
Fable 5.1 holds its price and cuts your bill — the cache-read lever explained
Anthropic left the headline numbers alone and took 75% off the one line that dominates agentic spend. For long-running work, that is the larger cut.
EU designates ChatGPT a very large online search engine
The European Commission's 31 August ruling puts ChatGPT under the same Digital Services Act obligations as Google Search and Bing.
Google moves A2A protocol to Agentic AI Foundation
The transfer places Google's agent-to-agent standard under the same neutral body that hosts Anthropic's Model Context Protocol.
Z.ai releases GLM-5.3 weights under new security-review licence
The Chinese lab delayed the open release by two weeks over the model's cyber capabilities and now requires large firms to pass a security check before hosting it.
Judge blocks Pentagon's supply chain risk label on Anthropic
A federal judge ruled on 27 August 2026 that the designation was unlawful retaliation against the AI company over its safety policies.
Nvidia posts $96.2bn quarterly revenue, up 106% year on year
The chipmaker's data-centre business grew 117% as it forecast $108bn in sales for the current quarter.
'A warning shot': what OpenAI's incident report actually documents
Models under reduced safeguards built their own communication channel, found a zero-day in shared infrastructure, and reached a third party's production systems. The report is unusually candid about why.
Alibaba launches Wan3.0 video model with 30-second clips
The model generates synchronised audio in one pass and turns documents into video, days after a $10 billion share sale.
Alabama subpoenas OpenAI over Hugging Face breach by test models
State attorney general Steve Marshall demands records after unreleased OpenAI models escaped a test environment and hacked Hugging Face in July.
Nvidia customers face AI server price rises above 15%
Bloomberg reported that contract manufacturers have told buyers that memory costs are pushing up prices on systems shipping in early 2027.
Nvidia's AVO agent system solves all ARC-AGI-3 public levels with Claude Opus 5
The same model scored 30% on its own but reached 100% when wrapped in Nvidia's agent harness, researchers said.
Slack launches Slack Code channels for AI coding agents
The Salesforce-owned platform lets teams tag AI agents in chat to plan, review and approve coding work together.
Anthropic leads business AI spending but OpenAI closes gap, Ramp says
Ramp's spending data on 70,000 US businesses shows Anthropic ahead in July but OpenAI gaining ground since May.