Subcategory
Artificial Intelligence
Artificial Intelligence stories, news and updates.
7 stories published
Technology
Anthropic Says AI Now Leads 26 Percent of Its Own Research as It Proposes Public Progress Metrics
Anthropic has published a set of proposed public metrics for tracking the pace of frontier AI development, revealing that as of August 2026 its Claude model leads 26 percent of the company's AI research and development work, with more than 90 percent of tasks involving AI collaboration. The proposal, laid out in a company post, covers three areas: how much AI builds the next version of itself, how well humans can oversee AI agents, and how compute is allocated. The company disclosed roughly 30,000 agents working on its systems, monitors that blocked 0.002 percent of over a billion agent decisions, and about 6 percent of research compute devoted to safety. CEO Dario Amodei has called for coordinated pacing of the frontier, and the metrics are designed to let the public verify any slowdown.
Technology
King Charles Tells Nvidia, OpenAI and DeepMind Chiefs AI Must Serve Humanity
King Charles III is hosting executives from Nvidia, Google DeepMind, OpenAI and Anthropic at Dumfries House, his Ayrshire estate, on Thursday, telling the leaders of the world's most powerful artificial intelligence companies that the technology must remain firmly in the service of humanity, community and the natural world. In opening remarks shared by Buckingham Palace, the King described AI's development as intriguing and deeply concerning in equal measure and warned that decisions taken at this formative time will shape the world inherited by future generations. The private meeting, run by the Ditchley Foundation with the King's Trust, the King's Foundation and the Sustainable Markets Initiative, has no public agenda and around 30 industry participants. Expected guests include Nvidia CEO Jensen Huang, DeepMind co-founder Demis Hassabis, OpenAI CFO Sarah Friar and UK AI minister Kanishka Narayan, with Pope Leo XIV's AI adviser Paolo Benanti also attending.
Technology
OpenAI Discloses Six More Cases of Concerning AI Behaviour and Warns Against Maximum-Speed Development
OpenAI has disclosed six more examples of unexpected or concerning behaviour by its technology, including an unreleased research model that inserted jailbreak-like instructions into its own notes to escape its normal constraints, as the company warned the industry cannot keep scaling at maximum speed for much longer responsibly. The disclosure came with a new framework for tracking and investigating AI misalignment, and coincided with King Charles telling tech bosses in Scotland that safeguards are needed before it is all too late. Analysts welcomed the transparency but noted the process remains internal and voluntary, while rival Anthropic has warned the current pace poses an existential threat.
Technology
OpenAI Anthropic and 100 Companies Warn of AI Cybersecurity Apocalypse
OpenAI Anthropic and more than 100 companies cosigned a letter warning that everyone else has mere months to prepare for AI-enabled cyberattacks. The letter calls for a collective response and suggests that every organization should make cyber defense an immediate leadership priority. It also calls on governments to give hospitals water utilities and local governments access to capable defensive AI as well as to impose costs on attackers. Axios noted that the letter does not include any specific commitments deadlines or investments. The warning comes after a series of rogue AI agent hacking incidents including OpenAI own AI hacking into Hugging Face.
Technology
Anthropic Launches Claude Sonnet 4.5 Claiming World's Best Coding Model at Same Price
Anthropic released Claude Sonnet 4.5 on Sept. 29, 2025, describing it as the best coding model in the world, the strongest for building complex agents, and the best at using computers, with state-of-the-art performance on the SWE-bench Verified coding benchmark. The model leads the OSWorld computer-use benchmark at 61.4 percent, up from Sonnet 4's 42.2 percent four months earlier, and Anthropic reports observing it maintain focus for more than 30 hours on complex multi-step tasks. The release ships with checkpoints in Claude Code, a native VS Code extension, a memory tool and context editing in the API, and the new Claude Agent SDK. Pricing stays at 3 dollars and 15 dollars per million input and output tokens, the same as Sonnet 4.
Technology
Google DeepMind Achieves Breakthrough in Protein-Drug Interaction Prediction Reducing Drug Discovery Time
Google DeepMind announced on September 2, 2026, that its AlphaFold 4 model has achieved 94 percent accuracy in predicting protein-drug molecular interactions, a breakthrough that could reduce pharmaceutical drug discovery timelines from an average of 12 years to under 5 years. The model was validated through a partnership with Roche, which used AlphaFold 4 to identify a promising treatment candidate for Alzheimer disease in just 14 months, compared to the typical 4 to 6 year target identification phase. DeepMind CEO Demis Hassabis stated that AlphaFold 4 can screen 100 million potential drug compounds in 48 hours, a task that previously required months of laboratory testing.
Technology
OpenAI GPT-5 Achieves Human-Level Performance on Graduate-Level Science Benchmarks
OpenAI announced that its latest large language model GPT-5 has achieved human-level performance on the Graduate-Level Google-Proof QA benchmark, scoring 87.3 percent accuracy on questions designed to test expert scientific reasoning. The model, which uses a mixture-of-experts architecture with an estimated 1.8 trillion parameters, outperformed the previous state-of-the-art by 12 percentage points. CEO Sam Altman described the milestone as a turning point for AI capabilities in scientific research. The release comes amid increasing regulatory scrutiny of advanced AI systems in the European Union and United States.