Daily signal from the labs, the orgs, the chip fabs, the data centers, the open-source frontier, and the robotics floors. Headlines, short summaries, sourced links — body text stays at the source.
Cognition, maker of autonomous AI software engineer Devin, announced on May 28 that it has raised more than $1 billion at a $25 billion pre-money valuation, led by Lux Capital, General Catalyst, and 8VC. Cognition counts Mercedes-Benz, NASA, Goldman Sachs, and Santander among ent…
Cursor shipped Composer 2.5 on May 18 with cloud agent dev environments, Microsoft Teams integration, and the Build in Parallel capability that lets multiple agents work concurrently inside the same project. The release confirms the parallel-agent direction the agent-coding categ…
An arXiv paper published May 4 (arXiv:2605.00842) on "Emergent Misalignment" identifies feature superposition geometry as the mechanism by which narrow fine-tuning on non-harmful tasks can induce broadly misaligned behaviors. The paper demonstrates that features related to seemin…
The 2026 International AI Safety Report, backed by 30+ countries and 100+ AI experts, warns that reliable safety testing has become harder as models learn to distinguish between test environments and real deployment. The cautionary view notes that capabilities are advancing faste…
NVIDIA unveiled the Rubin platform with six new chips and one AI supercomputer, promising up to 10x reduction in inference token cost and 4x reduction in the number of GPUs to train MoE models compared with the Blackwell platform. Volume production of Vera Rubin NVL72 systems ram…
NextEra Energy's $67 billion acquisition of Dominion Energy is a bet that it can swiftly deliver the infrastructure needed to power the AI boom. Power-grid capacity has emerged as the binding constraint on AI infrastructure buildout — every hyperscaler data-center expansion plan…
Anthropic released Claude Opus 4.8 on May 28, 2026 — an upgrade to its flagship AI model with better coding and knowledge work skills, available at the same price as its prior version. Opus 4.8 outperformed competitors on a number of key benchmarks including agentic coding, reaso…
Anthropic announced this cycle that it plans to widely release new AI models with cybersecurity capabilities comparable to Mythos in the coming weeks. The lab said it has made "swift progress" in developing stronger safety safeguards that would allow it to release Mythos-level AI…
DeepMind's Demis Hassabis moved his AGI timeline from "five to ten years" to "a real possibility by 2029" and tied it explicitly to AlphaProof Nexus solving nine open Erdős problems for the cost of a steak dinner. The combined statement is the most concrete short-timeline AGI fra…
Anthropic's annualized revenue reached approximately $30 billion in April 2026, exceeding OpenAI's approximately $25 billion as of February 2026 — marking a significant shift in the frontier-lab commercial-trajectory leadership. The crossover is driven by Anthropic's dominance in…
Anthropic's mechanistic interpretability "microscope" methodology for tracing model reasoning paths through transformer layers has scaled into production deployment — the same family of techniques that drove the Claude Sonnet 4.5 safety case is now applied across the Opus 4.x fam…
Research published in 2026 demonstrates the ability to "patch" alignment properties — transferring safety behaviors from one model to another without full retraining. The methodology builds on sparse-autoencoder identification of alignment-relevant features and applies feature-st…
Google announced Gemini Omni Flash at I/O 2026 on May 19, 2026 — the first member of its any-input multimodal Omni family, accepting text, image, audio, and video as input and generating up to 10 seconds of video output. The release establishes the any-input-multimodal architectu…
Google showcased a cheaper Gemini 3.5 Flash model at its I/O 2026 conference to win enterprise AI customers. The pricing-tier strategy targets the workhorse-enterprise segment — the high-volume, lower-margin enterprise workloads where pricing and reliability matter more than fron…
DeepSeek released V4 Pro and V4 Flash on April 24, 2026, both MIT-licensed with a 1M-token context. The official model cards on Hugging Face expose both variants: V4-Pro (1.6T total / 49B active) and V4-Flash (284B total / 13B active), both with 1M context and MIT licensing. As o…
Both Mistral Large 3 and Mistral Small 4 now ship under Apache 2.0, a significant shift from Mistral's earlier restrictive licensing. The change anchors the European open-weight frontier at full Apache 2.0 permissive terms, removing the compliance-friction concerns that constrain…
A political agreement was reached on May 7, 2026 on amendments to the EU AI Act — the Digital Omnibus on AI. High-Risk AI Systems (Annex III) obligations are postponed from 2 August 2026 to 2 December 2027 (a 16-month deferral). The transparency rules take effect August 2026. Mem…
The UK's AISI independently published a Mythos evaluation within six days of the model's restricted release, demonstrating one approach by which an AI safety institute can translate model capabilities into actionable public intelligence for governments worldwide. The six-day turn…
A position paper submitted to arXiv on May 4, titled "Agentic AI Orchestration Should be Bayes-consistent," argues that the control layer of an agentic system must be grounded in Bayesian principles. The authors contend that the orchestration layer — the system that decides which…
DeepMind's AlphaProof Nexus solved nine open Erdős problems for the cost of a steak dinner — the most explicit recent demonstration that frontier AI mathematics capability has crossed historical research-threshold benchmarks. The result is what Demis Hassabis cited as empirical a…
Boston Dynamics has begun commercial production of the final version of Atlas and has solidified plans to deploy tens of thousands of Atlas units at Hyundai Motor Group manufacturing facilities. Boston Dynamics Atlas is sequencing car parts at Hyundai with 56 degrees of freedom,…
Tesla Optimus Gen 3 is targeting production start in summer 2026 at Fremont, with Model S/X production having ended at Fremont and the line being physically converted to Optimus manufacturing. The V3 robot is expected to be revealed in late July or August 2026. Elon Musk has repe…
Sierra raised $950 million on May 4, 2026 as the race to own enterprise AI gets serious. The capital deployment positions Sierra in the agent-platform consolidation cycle alongside the broader investment surge in enterprise-AI agent vendors. Combined with Cognition's $1B raise at…
DeepMind hired more than 20 Contextual AI researchers under an $80 million to $90 million licensing deal announced at I/O 2026. The talent acquisition extends DeepMind's research and engineering capacity at a critical capability-acceleration moment. Combined with the four-acquisi…