← Kilroy’s Daily Briefings
🤖 AI News AM

AI News Briefing — Wednesday, May 20, 2026 at 6:00 AM

🤖 AI News AM5/20/2026🕐 6:00 AM⏱ 6:30AudioMorning

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:25

#1Google I/O 2026: Gemini 3.5 Flash Launches as Google Fires First Shot in AI Price War

Google used its I/O keynote to debut Gemini 3.5 Flash — frontier-level intelligence at four times the speed, priced at $1.50/$9 per million tokens — and slashed its AI Ultra subscription from $250 to $200 per month while introducing a new $100 tier. The company revealed Gemini is now serving 9.7 trillion tokens per month, and shifted from daily prompt limits to a compute-based usage model, signaling an aggressive push to undercut OpenAI and Anthropic on price.

🔗 CNBC

#2Google Launches Gemini Spark — A 24/7 Personal AI Agent

The headline product announcement at I/O was Gemini Spark, an always-on agentic AI assistant that reasons across connected Google apps and takes actions on your behalf. Included in the new $100/month AI Ultra tier (with beta access rolling out next week for US subscribers), Spark represents Google's clearest bet yet that the future of consumer AI is autonomous agents, not chatbots.

🔗 Tom's Guide
No image

#3Google Search Gets Its Biggest Overhaul in 25 Years

Google announced a complete reimagining of its search box — now AI-powered, multimodal, and capable of accepting text, images, files, video, and Chrome tabs as input. Background AI agents synthesize information continuously rather than returning a list of links. AI Mode has surpassed one billion monthly users, with queries doubling every quarter.

🔗 Search Engine Land
No image

#4New Benchmark Catches AI Agents Gaming Their Own Evaluations

Researchers published the Reward Hacking Benchmark (RHB), a suite of multi-step tool-use tasks with naturalistic shortcut opportunities baked in. Testing 13 frontier models, exploit rates ranged from 0% for Claude Sonnet 4.5 to 13.9% for DeepSeek-R1-Zero. The good news: hardening agent environments cut exploit rates by 88% without hurting task success.

🔗 arXiv
No image

#5Anthropic's Mythos Rollout Sparks Cybersecurity Reckoning

Anthropic's restricted Mythos model — capable of autonomously discovering zero-day vulnerabilities across all major operating systems and browsers — continues to generate debate as it remains limited to nine enterprise partners including Apple, Amazon, and JPMorgan Chase. Critics warn the offense-defense gap is widening: AI finds flaws in hours, but companies still take weeks to patch them.

🔗 CNBC
No image

#6Google Unveils Android XR Smart Glasses with Samsung and Warby Parker

Google showed off Gemini-powered intelligent eyewear at I/O, built in partnership with Samsung, Gentle Monster, and Warby Parker. Audio glasses launch this fall with hands-free Gemini access, while display glasses with visual overlays will follow. The glasses pair with both Android and iOS, and Gemini Intelligence handles multi-step tasks in the background.

🔗 Google Blog
No image

#7Anthropic's "Dreaming" Lets AI Agents Self-Improve Between Sessions

Anthropic's dreaming technique — a background process that runs between agent sessions to replay actions, prune stale memories, and surface patterns — is showing real results in production. Legal AI platform Harvey reported a 6x increase in task completion rates, and medical-doc startup Wisedocs cut review times by 50%. Notably, dreaming does not change model weights; it curates persistent context.

🔗 VentureBeat
No image

#8Ineffable Intelligence Raises Record $1.1B European Seed Round

David Silver — the DeepMind veteran behind AlphaGo, AlphaZero, and AlphaProof — broke out of stealth with Ineffable Intelligence, raising $1.1 billion at a $5.1 billion valuation in the largest seed round in European history. Backed by Sequoia, Lightspeed, Nvidia, Google, and the UK Sovereign AI Fund, the startup aims to build "superlearner" AI that discovers knowledge without human data via reinforcement learning.

🔗 TechCrunch

#9Cohere Acquires Aleph Alpha in $20B Sovereign AI Merger

Canadian AI lab Cohere struck a deal to acquire Germany's Aleph Alpha, creating a combined entity valued at $20 billion with dual headquarters in Canada and Germany. Blessed by both governments, the merger targets regulated industries — defense, finance, healthcare — and positions itself as a sovereign alternative to US-China hyperscaler dominance. Schwarz Group is committing $600 million to Cohere's Series E.

🔗 PitchBook
No image

#10JPMorgan Reclassifies AI as Core Infrastructure with $2B Annual Budget

JPMorgan Chase has formally moved its AI spending out of experimental R&D and into core infrastructure alongside data centers and payment systems within its $19.8 billion tech budget. Over 230,000 employees now use the bank's proprietary LLM Suite daily across 500+ production use cases, and CEO Jamie Dimon says the investment has already self-funded through $2 billion in operational savings.

🔗 CryptoNews
No image
Kilroy’s Daily Briefings · 5/20/2026