Nine developments in AI capability breakthroughs, governance collapse, supply chain attacks, and corporate restructuring.

On 20 May 2026, OpenAI announced that one of its internal general-purpose reasoning models autonomously disproved the Erdős unit distance conjecture, a central open problem in discrete geometry first posed by mathematician Paul Erdős in 1946. The proof – 125 pages – establishes an infinite family of planar point configurations that produce significantly more unit-distance pairs than the square-grid arrangements mathematicians had assumed to be optimal for nearly eight decades. Fields medalist Tim Gowers reviewed the proof independently and described it as "a milestone in AI mathematics." Princeton mathematician Will Sawin refined the result and quantified the improvement: the best configurations now scale as n^(1+δ), where δ ≥ 0.014, confirmed in a companion paper published the same day.

The method surprised the mathematical community most. Rather than extending known geometry techniques, the model connected the problem to algebraic number theory – specifically, Golod-Shafarevich class field towers – a deep branch of mathematics with no prior connection to combinatorial geometry. Number theorist Arul Shankar stated the work shows AI systems can move beyond assisting mathematicians to generating genuinely original ideas. OpenAI confirmed the model was not specifically trained on the unit distance problem and received only the problem statement before producing the proof independently.

Anthropic co-founder Jack Clark delivered the 2026 Cosmos Lecture at the University of Oxford on 20 May 2026, titled "Change is inevitable. Autonomy is not." He made four timed predictions: AI will collaborate with humans on a Nobel Prize-winning discovery within 12 months; bipedal robots will assist tradespeople within 2 years; companies run entirely by AI will generate millions of dollars in revenue within 18 months; and – most significantly – there is a 60%+ probability that by the end of 2028 an AI model will be able to fully train its successor, a process called recursive self-improvement. Clark used the phrase "intelligence explosion" – long confined to AI safety theory – and confirmed it now appears in official Anthropic research documents. He also stated there is a "non-zero chance" AI could kill everyone on the planet, and that this risk "hasn't gone away."

The Anthropic Institute, Clark's research division on the societal consequences of advanced AI, published a five-page research agenda this week explicitly referencing the intelligence explosion framing. Clark's 60% estimate is not a detached forecast; it describes the trajectory of work Anthropic is actively pursuing, particularly following the company's hiring of Andrej Karpathy specifically to use Claude to accelerate pretraining research – which is, in practice, early-stage recursive self-improvement.

A leaked audio recording from a Meta all-hands meeting on 30 April 2026, published by labor-focused outlet More Perfect Union on 19 May, surfaced publicly on the same day approximately 8,000 Meta employees received layoff notices. In the audio, CEO Mark Zuckerberg describes a programme called the Model Capability Initiative (MCI): an internal effort that monitored employee activity on Gmail, Google Chat, the internal assistant Metamate, and VS Code to train Meta's AI models on how highly capable people work. "The AI models learn from watching really smart people do things," Zuckerberg says in the recording. He added: "No human is looking at or watching what people are doing on their computers."

The reaction across Meta's workforce was immediate and organized. Employees pasted protest flyers inside meeting rooms and circulated a petition against MCI. The consent and transparency question here is not whether the tracking was technically anonymous – Zuckerberg's assurance on that point may be accurate. The question is whether harvesting work patterns without explicit, informed consent for a purpose directly connected to reducing the roles of the people being monitored is ethically defensible, irrespective of anonymization. No independent verification of Meta's anonymization claims has been published.

The White House AI executive order – which would have established a voluntary 90-day pre-launch review framework for frontier AI models, with NSA involvement in classified capability testing – was cancelled on 21 May 2026, hours before the scheduled signing ceremony. Invitations to the ceremony had already been distributed. Axios reported that the primary reason was that Trump "just hates regulation" and that former AI czar David Sacks "hated it" too, describing the order as "unnecessary" and "just something doomers wanted." Meta CEO Mark Zuckerberg, xAI CEO Elon Musk, and Sacks all spoke with Trump directly between Wednesday night and Thursday morning. By the time the cancellation was announced, some executives invited to the signing were already on planes.

Trump explained the decision publicly: "I didn't like certain aspects of it. I postponed it. I think it gets in the way of – you know, we're leading China, we're leading everybody, and I didn't want to do anything to get in the way of that lead." The draft order was partly motivated by Anthropic's Claude Mythos discovering over 10,000 critical software vulnerabilities at scale – a concrete, documented security risk that the interagency process had been building around. That evidence base was overridden by informal executive access.

GitHub confirmed on 21 May 2026 that approximately 3,800 internal repositories were exfiltrated by threat actor group TeamPCP (also tracked as UNC6780) following the compromise of an employee device. The attack vector was a trojanized version of the Nx Console VS Code extension – which has 2.2 million installs and verified publisher status – that was live on the Visual Studio Marketplace for exactly 18 minutes, between 12:30 PM and 12:48 PM UTC on 18 May 2026. During that window, the malicious extension ran silently on startup, executing a shell command disguised as a routine MCP setup task that downloaded a credential stealer from a planted commit in the official nrwl/nx GitHub repository. The stealer targeted 1Password vaults, Anthropic Claude Code configuration files (at ~/.claude/settings.json), npm tokens, GitHub personal access tokens, and AWS credentials.

GitHub CISO Alexis Wales confirmed: "We have no evidence of impact to customer information stored outside of GitHub's internal repositories." The 18-minute window between a trojanized extension being published and removed is the defining operational fact of this incident. Standard endpoint security does not monitor VS Code extension installs in real time. Auto-update, enabled by default across VS Code, gives any publisher with a compromised release pipeline a direct push channel to every developer machine – with no marketplace review gate or waiting period.

The GitHub breach is one confirmed stop on a broader campaign. TeamPCP's Mini Shai-Hulud worm (Wave 4) launched on 11 May 2026 by compromising TanStack's entire router ecosystem, spreading across 170+ npm packages and two PyPI packages in a single coordinated campaign. The associated CVE, CVE-2026-45321, carries a CVSS score of 9.6. Confirmed victims beyond GitHub include: OpenAI (two employee devices compromised; limited credential material exfiltrated from internal source code repositories; iOS, macOS, and Windows code-signing certificates rotated); Mistral AI (one developer device compromised, facing a $25,000 Monero extortion demand for an alleged 5 GB source code leak); the European Commission's public website; and data contracting firm Mercor. Trend Micro has tracked at least seven confirmed TeamPCP waves in 2026 alone, with prior targets including Aqua's Trivy security scanner, Checkmarx KICS, LiteLLM, and Bitwarden CLI.

OpenAI confirmed the certificate rotation timeline publicly: iOS and Windows certificates have already been rotated; the macOS ChatGPT desktop application code-signing certificate will be fully revoked on 12 June 2026. This means the current macOS ChatGPT desktop app will become untrusted by macOS Gatekeeper after that date – users will need to reinstall the application from a freshly signed build. OpenAI has not yet issued a public notification to end users about this requirement. TeamPCP is building a self-sustaining attack loop: compromise one trusted developer tool, harvest credentials, use those credentials to access the next trusted tool or repository, repeat.

Google announced at and immediately following I/O 2026 (19 – 21 May) that three major creative platforms are integrating natively into the Gemini app: Adobe (via the Firefly AI Assistant agent, providing access to Photoshop, Premiere Pro, Illustrator, Lightroom, and Express from within a Gemini conversation), Canva (via "Magic Layers," enabling users to generate an image in Gemini and unlock it in Canva with every element on a separate editable layer – already in early rollout for Gemini AI Ultra subscribers), and CapCut (announced 21 May, described as coming soon). CapCut has been downloaded over 1.2 billion times globally. Adobe CEO Shantanu Narayen described the partnership as combining "Adobe's creative DNA with Google's AI models to usher in a new era of creative expression."

The Canva Magic Layers integration is the most fully deployed. Users generate an image in Gemini and the resulting file arrives in Canva with each generated element (text, objects, backgrounds) on independent, editable layers – eliminating the rasterized, single-layer limitation that previously made AI-generated images difficult to refine professionally. Google also confirmed Gemini AI Overviews now reaches over 2.5 billion monthly users; Gemini's conversational AI Mode has exceeded 1 billion users, with broader rollout scheduled for this summer.

Anthropic confirmed on 18 May 2026 that it has acquired Stainless, a startup founded in 2022 that builds high-quality software development kits for API products. Stainless has powered the generation of every official Anthropic SDK since the earliest days of the Claude API. It also built the SDKs used by OpenAI, Cloudflare, Merge, and hundreds of other API-driven companies. The acquisition is described by Anthropic as extending Claude's ability to connect to data and tools as the frontier of AI shifts from models to agents. Katelyn Lesse, Anthropic's Head of Platform Engineering, stated: "Agents are only as useful as what they can connect to. We're excited to bring the Stainless team into Anthropic to advance Claude's ability to connect to data and tools." Alex Rattray, founder and CEO of Stainless, confirmed the acquisition: "I started Stainless because SDKs deserve as much care as the APIs they wrap."

Stainless's approach generates SDKs from OpenAPI specifications with consistent patterns, strong typing, and idiomatic language conventions in TypeScript, Python, Go, Java, and more. The acquisition brings in-house the team responsible for how developers experience the Claude API at the code level – including CLIs and MCP server tooling. By acquiring the leading SDK generation tooling for MCP servers, Anthropic now controls both the protocol and the primary implementation layer. No acquisition price was disclosed. For developers who have observed the gap in ergonomics between Claude's SDKs and OpenAI's SDKs, that gap is now being closed deliberately and in-house.

Intuit announced on 20 May 2026 that it is cutting approximately 3,000 jobs – roughly 17% of its 18,200-person global workforce – in a broad restructuring framed around reducing organizational complexity and accelerating AI integration across TurboTax, QuickBooks, Credit Karma, and Mailchimp. CEO Sasan Goodarzi communicated the decision via an internal memo reported by Reuters: "We believe we can serve more customers and deliver breakthrough products that fuel our customers' success by reducing complexity and simplifying our structure to become a faster, leaner, and more focused company." Affected US employees will remain with the company until 31 July 2026. The company will recognize the bulk of restructuring charges during its fourth fiscal quarter ending July 2026. Separately, Reuters reported that Intuit has signed multi-year AI partnership agreements with both Anthropic and OpenAI – structured to embed those companies' models into Intuit's products while also making Intuit's tax, accounting, financial, and marketing tools available inside Claude and ChatGPT respectively.

Intuit had approximately 18,200 employees worldwide as of July 2025 per its annual report, making this reduction roughly 3,100 positions by headcount. The restructuring includes closing two offices and scaling back Mailchimp operations. The company merged TurboTax and Credit Karma operations in the prior year, identifying redundant positions that contributed to the total. Intuit's CEO compensation was reported at $36.8 million including cash incentives and stock awards for fiscal 2025. The pattern – strong earnings quarter, large workforce reduction, AI infrastructure investment, new AI partnership agreements – is now a standard corporate playbook in the enterprise software sector.