Anthropic launched "dreaming," a new capability that lets Claude Managed Agents review past sessions, distill learnings into plain-text notes and structured playbooks, and self-improve over time — without modifying model weights. Alongside dreaming (now in research preview), the company moved outcomes (rubric-based self-evaluation) and multi-agent orchestration (parallel specialist delegation) into public beta. Early adopter Harvey reports a 6x increase in task completion rates; Netflix is processing logs from hundreds of builds simultaneously with multi-agent orchestration.
Anthropic secured access to SpaceX's entire Colossus 1 data center in Memphis — over 220,000 NVIDIA GPUs and 300+ MW of power — to meet exploding Claude demand. The company reported 80x year-over-year growth in revenue and usage for Q1 2026. Paid Claude Code users immediately get doubled rate limits, with peak-hour reductions eliminated and Opus API limits raised.
OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as ChatGPT's default, delivering 52.5% fewer hallucinations on high-stakes prompts and 37.3% fewer inaccurate claims on user-flagged conversations. The model is roughly 30% more concise and cuts back on "gratuitous emojis." A new memory-source transparency feature shows which prior context shaped responses, rolling out first to Plus and Pro users.
Four-person startup Subquadratic launched SubQ, claiming the first frontier LLM that fully escapes quadratic attention scaling. Using a novel Subquadratic Sparse Attention mechanism, SubQ supports a 12-million-token context window in research mode and claims ~1,000x attention compute reduction at that scale. Researchers are demanding independent verification — weights are closed, the full technical report is unreleased, and prior subquadratic architectures have repeatedly underperformed transformers at frontier scale.
IREN Limited signed a five-year contract to provide NVIDIA with managed GPU cloud services for its internal AI and research workloads, deployed across ~60 MW at IREN's Childress, Texas campus using air-cooled Blackwell systems. The deal includes a strategic 5 GW partnership and rights for NVIDIA to purchase up to 30 million IREN shares (~$2.1B) at $70/share. IREN shares jumped as much as 27% on the news.
The Department of War finalized agreements with AWS, Google, Microsoft, NVIDIA, OpenAI, Oracle, Reflection, and SpaceX to deploy AI on IL-6 (Secret) and IL-7 (Top Secret) classified networks. Anthropic was conspicuously excluded after refusing terms that would allow military use of Claude for autonomous weapons and mass surveillance. A federal judge has issued an injunction blocking subsequent Trump administration orders to sever all federal ties with Anthropic.
Washington and Beijing are negotiating first-ever formal AI talks ahead of the May 14-15 Trump-Xi summit in Beijing, aiming to prevent AI competition from escalating into military or security crises. The proposed dialogue would address risks from unpredictable AI models, autonomous weapons, and nonstate actors leveraging open-source AI. Treasury Secretary Scott Bessent is leading the US side; China has yet to name its counterpart.
Microsoft's Q1 2026 AI Diffusion Report shows global AI usage rose from 16.3% to 17.8% of the world's working-age population, with 26 economies now above 30%. The UAE leads at 70.1%, the US climbed to 21st at 31.3%, and Asian adoption is accelerating — led by South Korea, Thailand, and Japan. But the divide is growing: 27.5% adoption in developed nations versus 15.4% in developing ones. Notably, US software developer employment hit a record 2.2 million, up 8.5% YoY.
Connecticut's bipartisan SB5 — a 67-page omnibus covering consumer disclosures, frontier-developer safety obligations, whistleblower protections, and minors-specific chatbot safeguards — is headed to Governor Lamont's desk. Iowa Governor Kim Reynolds signed SF 2417, which mandates AI disclosures, bans manipulative chatbot engagement tactics targeting minors, and prohibits AI from presenting itself as a licensed mental health provider. These join laws in New York, California, and five other states.
The Center for AI Standards and Innovation (CAISI) announced agreements with Google DeepMind, Microsoft, and Elon Musk's xAI to evaluate their AI models before they're publicly available, building on earlier partnerships with OpenAI and Anthropic. This represents the most concrete step yet toward pre-release government AI safety testing under the current administration, though details on evaluation criteria and enforcement mechanisms remain thin.
