Live Feed
Unit 42's 'Zealot' Demonstrates AI Cloud Attacks Have Reached Functional Maturity — SSRF to Exfiltration in Minutes
Palo Alto Networks Unit 42 built and tested an autonomous multi-agent system that chains server-side request forgery, credential theft, and BigQuery exfiltration against misconfigured GCP with minimal human guidance. No zero-days required.
AMD Data Center Hits $5.8B in Q1 2026, Up 57% as EPYC and Instinct GPU Ramp Accelerates
AMD reported $10.3B in Q1 2026 revenue with Data Center at $5.8B, up 57% year-over-year. EPYC processors and Instinct GPU shipments are now the primary drivers of revenue and earnings growth.
Google, Microsoft, and xAI Agree to CAISI Pre-Release AI Testing — Government Gets Guardrail-Off Access Before Launch
The US Commerce Department's CAISI unit will test frontier AI models from Google DeepMind, Microsoft, and xAI before public release, including versions with reduced safety filters. OpenAI and Anthropic signed equivalent agreements in 2024. Anthropic's Claude Mythos triggered the policy acceleration.
SubQ Launches 12M-Token Model on Sub-Quadratic Sparse Attention — 52x Faster Than FlashAttention at 1M Tokens
SubQ has launched what it claims is the first frontier-scale model built entirely on sub-quadratic sparse attention architecture. At 1M tokens it runs 52x faster than FlashAttention and costs less than 5% of Claude Opus. Early access is live at subq.ai.
Coinbase Cuts 14%, PayPal Cuts 20%: Fintech Is Restructuring Around AI Agents — Not Just Cutting Costs
Coinbase is eliminating 14% of its workforce and rebuilding around 'one-person product teams' that combine engineering, design, and product work. PayPal is cutting 20% (~4,500 jobs) with a $1.5B savings target. Both CEOs explicitly cite AI as the structural driver, not just a cost tool.
Zuckerberg Personally Named in Publishers' Copyright Lawsuit Over Llama Training Data
Five publishing houses and author Scott Turow filed a federal class action in Manhattan accusing Meta and CEO Mark Zuckerberg of using millions of copyrighted books to train Llama. The suit alleges Zuckerberg personally authorized and actively encouraged the infringement.
GPT-5.5 Instant Is Now the Default ChatGPT Model — 52.5% Fewer Hallucinations, Gmail Context, $5/M Input
OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as the default ChatGPT model. The update cuts hallucinated claims by 52.5% on high-stakes prompts, trims output length by 30%, and adds memory integration across past chats, files, and connected Gmail.
Anthropic Commits $200B to Google Cloud Over Five Years — Enough to Anchor 40% of Google's Backlog
Anthropic has agreed to spend $200 billion on Google Cloud and TPU chips over five years, per The Information and Reuters. The commitment may represent more than 40% of Google's recently disclosed revenue backlog, making a single AI lab one of Google Cloud's largest contracted customers.
Anthropic Launches 10 Finance Agent Templates — From KYC Screening to Month-End Close, Powered by Opus 4.7
Anthropic released ten ready-to-run agent templates covering every major financial services workflow, shipping as plugins in Claude Cowork and as cookbooks for Claude Managed Agents. Claude Opus 4.7 leads Vals AI's Finance Agent benchmark at 64.37%.
Computer Use Is 45x More Expensive Than Structured APIs — and Gets the Task Wrong
Reflex.dev benchmarked a vision agent against an API agent on the same admin panel task. The vision agent missed three of four target records and cost 45 times more. The API agent finished in 8 calls with full accuracy.
Google Chrome Pushed a 4GB On-Device AI Model to Millions of Users Without Consent — and It Restores Itself
Chrome silently installed Gemini Nano as a 4GB weights.bin file in user profiles, powering browser AI features without any opt-in dialog. Deleting it triggers automatic restoration. The only durable block requires a Windows Registry policy edit that most users will never find.
Amazon Opens Claude Code and Codex to All Engineers After 1,500-Strong Internal Petition
Amazon VP Jim Haughwout confirmed company-wide access to Claude Code effective immediately and Codex on May 12. Both run inside Amazon Bedrock. Engineers had been blocked from production use while simultaneously selling the same tool to external customers.
Anthropic Forms $1.5B Joint Venture With Blackstone, Goldman Sachs, and Hellman & Friedman to Capture PE Portfolio AI Spend
Anthropic, Blackstone, and Hellman & Friedman are each putting in roughly $300M, Goldman Sachs $150M, to create an AI deployment company targeting private equity-owned businesses. The JV will not just resell Claude access — it will rebuild workflows.
Microsoft's AI Business Hits $37B ARR at 123% Growth — Capex Rises to $190B for 2026
Q3 FY2026 results show Microsoft's AI revenue run rate at $37 billion, up 123% year-on-year. The company raised its 2026 capex plan to $190 billion, with $25 billion of that attributable to hardware cost inflation alone.
GPT-5.5 Costs 49-92% More in Practice — OpenRouter's Switcher Cohort Study Breaks Down the Real Price
OpenRouter analyzed users who moved from GPT-5.4 to GPT-5.5 after launch. Net cost increases run from 49% on long-context work to 92% on short prompts. The model's verbosity reduction only kicks in above 10K input tokens.