President Trump signed an executive order on June 2 asking AI companies to give the government access to their most powerful models up to 30 days before release. Two weeks after scrapping a tougher draft, the rewrite is explicitly voluntary, with no licensing or preclearance requirement.
At Computex, Jensen Huang unveiled the RTX Spark Superchip: 20 Arm cores, a Blackwell GPU, and 128GB of unified memory in a Windows laptop Nvidia says can run a 120-billion-parameter model with a million-token context entirely on-device. Over 30 laptops and 10 desktops ship this fall.
As of June 1, every GitHub Copilot plan bills by the token instead of a flat fee, using a new currency called AI Credits. Base prices held steady, but power users running agentic sessions are posting projected cost increases of 10x to 50x, and some are cancelling.
Anthropic raised $65 billion at a $965 billion valuation, passing OpenAI for the first time. The round is backed by a wave of new and existing investors and tied closely to a multi-gigawatt compute buildout with Amazon, Google, and SpaceX.
Claude Opus 4.8 leads agentic coding benchmarks and is markedly more honest about its own work. The bigger change for engineers is dynamic workflows, a new Claude Code feature that fans out hundreds of parallel subagents to run whole-codebase migrations.
Anthropic published its first Project Glasswing update on May 22, 2026, reporting more than 10,000 high- or critical-severity vulnerabilities found by Claude Mythos Preview in a single month. The findings include a critical wolfSSL flaw (CVE-2026-5194) affecting roughly 5 billion devices. Open-source maintainers have asked Anthropic to slow disclosures while they catch up on patches.
Cognition closed a Series D round of more than one billion dollars on May 27, 2026, at a $26 billion post-money valuation. The startup behind the autonomous coding agent Devin reported $492 million in annualized revenue, 50% month-over-month growth, and 89% of its own code now written by Devin itself.
If signed, an Anthropic-Microsoft Maia 200 deal would resolve a two-year question for Microsoft's silicon program: whether anyone outside the company will pay to use it. It would also make Maia 200 Anthropic's fourth custom silicon partner.
Iranian state-sponsored hackers ran three campaign waves during and after Operation Epic Fury, including a new MiniFast backdoor delivered through fake Oracle SQL Developer downloads. Check Point Research says the code carries fingerprints of AI-assisted development.
Anthropic is on track for its first quarterly operating profit ever, projecting 10.9 billion dollars in Q2 revenue and 559 million in operating profit. Compute cost per revenue dollar fell from 71 cents to 56 cents in one quarter, a sharp reversal from the company's August 2025 guidance of no profits until 2028.
TrapDoor is the first widely-reported supply chain campaign that weaponizes AI coding assistants. Beyond credential theft via 384 versions across three package registries, the attacker hides instructions in CLAUDE.md and .cursorrules using zero-width Unicode, hoping the AI assistant runs commands the developer cannot see.
A backdoored Nx Console extension was live on the Visual Studio Marketplace for just 18 minutes on May 18, but it was enough for the group TeamPCP to steal credentials at GitHub, OpenAI, Grafana Labs, and Mistral AI and exfiltrate roughly 3,800 of GitHub's internal repositories.
At Oxford on May 20, Anthropic co-founder Jack Clark predicted AI would help win a Nobel Prize within 12 months and gave 60%-plus odds that an AI model could fully train its own successor by the end of 2028. He also repeated that the technology carries a non-zero chance of catastrophe.
OpenAI is preparing to confidentially file a draft IPO prospectus with the SEC as soon as Friday, May 22, 2026, with Goldman Sachs and Morgan Stanley steering a public debut targeted for September at a valuation that could top $1 trillion. The move follows October 2025's conversion to a Public Benefit Corporation and February's record $122 billion round at an $852 billion valuation. CFO Sarah Friar has reportedly pushed back on the aggressive timeline, citing roughly $600 billion in five-year compute commitments against an annualized revenue run rate near $25 billion.
On May 20, 2026, SpaceX's S-1 filing disclosed that Anthropic will pay about $1.25 billion a month, more than $40 billion over three years, to rent the full 300-megawatt output of xAI's Colossus 1 data center near Memphis. The filing also revealed xAI lost $6.4 billion from operations in 2025 on $3.2 billion in revenue, and that Grok's daily active users fell from 13.9 million to 12.2 million between March and April. The deal is a textbook neocloud move: building AI infrastructure, then renting the idle capacity to a rival.
On May 20, 2026, OpenAI said a general-purpose reasoning model produced an original proof disproving Paul Erdős's planar unit distance conjecture, open since 1946. The model reached into algebraic number theory to find an infinite family of point arrangements beating the square grid. Unlike OpenAI's debunked October claim, this result was reviewed by mathematicians including Fields Medalist Tim Gowers and Thomas Bloom, the skeptic who exposed the earlier mistake.
President Trump scrapped the signing of a landmark AI executive order on May 21, 2026, telling reporters he did not want to blunt America's lead over China. The order, in development for weeks after Anthropic's Mythos model autonomously found thousands of cyber vulnerabilities, would have created a voluntary framework for labs to share frontier models with the government 90 days before release. Even that compromise proved too much for an administration torn between China competition and AI safety.
Google launched Gemini 3.5 Flash at I/O 2026, a fast, cheap model that outperforms its flagship Gemini 3.1 Pro on nearly every benchmark. The release, paired with Antigravity 2.0 and Managed Agents in the Gemini API, marks Google's strategic shift from conversational AI to autonomous agents.
