AI News
Our editorial picks on the AI stories that matter for Builders.
Gemini 4 Argon: Google's Cyber-Defence Model
Google's first Gemini 4 model, Argon, goes first to vetted cyber defenders, and Google claims a 1M-token output limit. What is confirmed, what is claimed, and what UK and Indian teams should do.
UK AI Took $12.6B of $17B in H1 2026: What Builders Should Know
UK startups raised $17B in H1 2026 and AI took $12.6B, according to HSBC Innovation Banking and Dealroom. Here is what the figures mean and why trackers disagree.
Claude Sonnet 5.5: Same $2/$10 Price, 30% Faster, Fewer Tokens
Claude Sonnet 5.5 keeps the $2/$10 list price, generates output 30%+ faster and, Anthropic says, costs up to 30% less per task. How to test that, and when Opus 5.5 earns 2x.
OpenAI DevDay: GPT-6.1 Sol, Dots Agents and Codex Cloud for Builders
GPT-6.1 Sol keeps $2/$10 but halves cached input, Dots bring always-on Astra agents, and Ultrafast costs 6x. What OpenAI's DevDay changes for agent stacks in India and the UK.
Agents Can Delete Their Own Traces: What That Means for Audit Logs
An arXiv paper found that, in full-access mode, most tested coding-agent harnesses let agents delete their own session traces when asked. How to make logs tamper-evident.
ProgramDistill Mines 4,063 Coding Tasks From 26 Live Web Apps
KAIST and Microsoft built a pipeline that mines replay-verified behaviours from live web apps and turns them into 4,063 coding-agent tasks with no human labelling.