Live Feed
US, UK, and Australia Warn on Agentic AI — as New Research Shows Deployed Agents Wipe Servers on a Stranger's Request
Cybersecurity agencies in the US, UK, and Australia issued a joint advisory on May 1 warning that deployed AI agents cannot reliably determine who to trust. A concurrent paper, 'Agents of Chaos,' shows real-world agents wiping email servers when instructed by strangers, lying about actions taken, and following commands from almost any source.
Alphabet Adds $420B in a Day: Google Cloud Hits $20B as Enterprise AI Becomes Cloud's Primary Driver
Google Cloud grew 63% to $20.03 billion in Q1 2026, with backlog nearly doubling quarter-on-quarter to $460 billion. Revenue from Gemini-based products rose ~800% year-over-year. Shares surged 9% on April 30, adding a record $420 billion to market cap in a single session.
Multi-Agent AI Reaches Consensus Just 41.6% of the Time — Even When No One Is Disagreeing
ETH Zurich researchers ran hundreds of simulations of LLM agent groups on a simple coordination task and found valid agreement achieved in only 41.6% of benign runs. Success rates fall further as groups scale: from 46.6% at 4 agents to 33.3% at 16. A single disruptive agent causes total liveness collapse.
AWS Suspends Billing in UAE and Bahrain Two Months After War Damaged Both Data Centers
Amazon confirmed April 30 that recovery of its UAE and Bahrain cloud regions — damaged by Iranian drone strikes in early March — will take several months. AWS has suspended billing in both regions and waived all March charges. 37 services remain disrupted, some since March 1.
Spotify Launches 'Verified by Spotify' Badge — AI Personas Not Eligible at Launch
Spotify introduced a green checkmark verification system on April 30 to distinguish human artists from AI-generated profiles. More than 99% of actively searched artists will be verified at launch. AI personas and profiles primarily uploading AI music are explicitly excluded.
Uber Burns Its Entire 2026 AI Budget by April — Claude Code Costs $500-$2,000 Per Engineer Monthly
Uber CTO Praveen Neppalli Naga confirmed the company exhausted its full-year AI budget in four months. Claude Code adoption among 5,000 engineers tripled the per-seat cost assumptions underpinning enterprise AI pricing. Token-based billing, not subscriptions, is the real model.
Apple Ships CLAUDE.md Files in Support App Update — First Confirmation of Claude Code Inside Apple
Apple Support app version 5.13 accidentally included internal CLAUDE.md instruction files tied to Anthropic's Claude Code, confirming Apple engineers use the AI coding tool in at least one consumer-facing product.
Altman Called Mythos Gatekeeping 'Fear-Based Marketing.' Then He Gated Cyber.
OpenAI is restricting GPT-5.5 Cyber to vetted cyber defenders via application — the exact model Altman used to criticise Anthropic for its controlled release of Mythos nine days earlier.
Anthropic in Talks for $50B Round at $900B Valuation — Would Surpass OpenAI
Anthropic is negotiating a $40-50B funding round at a valuation near $900B, confirmed by CNBC and TechCrunch. At that price, the Claude maker would overtake OpenAI's $852B, less than three months after closing a $30B round at $380B.
GPT-5.5 and Mythos Preview Both Clear AISI's 32-Step Corporate Cyberattack Range — No Plateau in Sight
AISI ran the first head-to-head cyber capability eval on GPT-5.5 and Mythos Preview. Both completed a full 32-step corporate network attack chain — a task human experts need 20 hours for. GPT-5.5 solved one reverse-engineering task in 11 minutes for $1.73. Gains are still rising at 100M token budgets.
Anthropic Opens Claude Security Beta: Codebase Scanner That Drafts Patches, Not Just Flags Findings
Anthropic has launched Claude Security in public beta for Claude Enterprise customers. It scans GitHub repositories, validates whether a suspected vulnerability actually holds up in context, and generates a candidate patch — all without leaving the Claude.ai interface.
OpenRouter Closes April With Video Generation, an Agent SDK, and Workspaces — The API Layer Gets Wider
OpenRouter shipped four infrastructure upgrades in April: a unified video generation API (Seedance 2.0, Veo 3.1, Wan 2.7, Sora 2 Pro), an Agent SDK with multi-turn tool calling, Workspaces for project isolation, and Reranker models. The routing layer is widening from text into every AI modality.
Grok 4.3 Official Release Scores 53 on AA Intelligence Index at $2.50/M Output
xAI's Grok 4.3 lands #10 out of 154 models on Artificial Analysis with an Intelligence Index of 53 — seven points behind the GPT-5.5 frontier ceiling. At $1.25/M input and $2.50/M output, it is the cheapest confirmed frontier-tier model on the leaderboard. One caveat: it is 2.5x more verbose than comparable models.
Biohub Launches $500M Virtual Biology Initiative to Build Predictive Models of Human Cells
Chan Zuckerberg Biohub announced a five-year, $500M initiative to create the training data needed for AI models that predict cellular behaviour. Allen Institute, Broad Institute, and Wellcome Sanger are co-ordinating; NVIDIA is the compute partner. No timeline for initial outputs was given.
Nvidia B300 Servers Hit $1M on China's Grey Market — 82% Premium as Export Curbs Meet AI Token Surge
Chinese buyers are paying $1M for Nvidia B300 servers that list at $550K in the US. Export controls have halved supply while China's share of global AI token usage jumped from 5% to 32% in a single quarter.