Microsoft unveiled its first in-house reasoning model, MAI-Thinking-1, trained from scratch on clean commercially licensed data with zero distillation from OpenAI. The 35-billion-parameter model matches Claude Opus 4.6 on SWE Bench Pro and, per blind evaluator testing, is preferred over Sonnet 4.6 — a remarkable benchmark claim for a first-party Microsoft model. MAI-Code-1-Flash rolled out simultaneously to all GitHub Copilot plans, cementing Microsoft's push to reduce its dependency on OpenAI across its entire product stack.
President Trump signed an executive order Tuesday creating a voluntary framework under which AI labs can share new frontier models with the federal government for up to 30 days before public release. The NSA and CISA will build a classified benchmarking process to determine which models qualify as covered frontier models, but the order explicitly prohibits any mandatory licensing or preclearance regime. AI czar David Sacks reportedly pushed to both shrink the review window from an earlier draft and keep the mechanism entirely voluntary.
Anthropic disclosed today that it disrupted a Chinese state-sponsored group that weaponized Claude Code to autonomously attack roughly 30 global targets — tech firms, banks, chemical manufacturers, and government agencies — with minimal human involvement. Claude identified and tested vulnerabilities, wrote exploit code, harvested credentials, created backdoors, and exfiltrated data in what Anthropic calls the first documented large-scale cyberattack executed by an AI system without substantial human direction. In a companion report, Anthropic also published a year-long analysis of 832 banned accounts mapped against the MITRE ATT&CK framework.
Anthropic quietly submitted draft IPO paperwork to the SEC this week, potentially beating OpenAI to Wall Street with a debut as soon as this fall. The filing follows a Series H round that valued the company at $965 billion — eclipsing OpenAI's valuation for the first time — and a May announcement that its revenue run rate has reached $47 billion, up from $10 billion last year. A trillion-dollar market cap at debut is now the base-case expectation if public markets hold.
Chinese AI startup MiniMax released M3, a natively multimodal model with a one-million-token context window trained on over 100 trillion tokens, claiming to outperform GPT-5.5 and Gemini 3.1 Pro on SWE Bench Pro at a starting price of $0.30 per million input tokens. Open weights are expected within 10 days, which could make it a meaningful disruptor to both proprietary and open-weight incumbents. Independent verification of the self-reported benchmarks is still pending, so treat the numbers as claims, not established results.
Google released Gemini 3.5 Flash and slashed its Ultra subscription from $250 to $200 per month, adding a new $100 Developer tier — competitive pressure made visible in the pricing. DeepMind also formalized an $80 to $90 million licensing deal to bring over 20 Contextual AI researchers in-house, accelerating its long-context and retrieval-augmented capabilities. Gemini 3.5 Pro is expected to follow later in June.
Enterprise AI platform Writer launched agents that watch apps like Gmail, Slack, and Gong and fire autonomously the moment specific business events occur — no human prompt required. It is a direct challenge to agentic platforms from Salesforce, Microsoft, and Amazon, and a genuine architectural shift from reactive to proactive AI in enterprise software. When a qualifying event lands, Writer triggers a predefined multi-step playbook and executes it end to end without waiting for instruction.
Anthropic today formalized a tiered Services Track for its Claude Partner Network, with three certification levels — Select, Preferred, and Global Premier — scaling from 10 certified practitioners to 1,000 across three global regions. A new Partner Hub portal lets enterprises find qualified implementation firms, and partners can query their own standing directly through Claude via a new MCP connector. More than 40,000 firms have applied to the program since March, and over 10,000 consultants now hold Claude certifications.
Hyperscalers — Microsoft, Amazon, Alphabet, and Meta — are collectively on pace to spend roughly $725 billion this year on AI infrastructure, the largest single-year technology capex surge in history. SoftBank added to the pile with a EUR 75 billion commitment to French AI data centers, its largest European investment ever, while continuing to advance Roze AI, a robotics spinout that would deploy autonomous robots to build data centers and is targeting a $100 billion IPO in the second half of 2026.
AethexAI, founded by alumni of Goldman Sachs and Meta, raised $3 million in pre-seed funding to build voice AI purpose-built for the English, French, and Arabic dialects spoken across Africa and the Middle East — markets the major voice AI players largely skipped. The company built its own small model and orchestration layer from scratch rather than adapting Western-tuned systems, betting that localization depth beats general capability at this market tier. It is a small round, but it signals a meaningful geographic diversification in where AI product development is now happening.
