What Exactly Happened

According to the official Claude Status Page, the system-wide outage began on June 23 at 14:19 UTC (16:19 Central European Time). Errors manifested as elevated error rates across all models — from Opus 4.8 to Sonnet 4.6 to Haiku 4.5. Anthropic's engineers identified the cause within six minutes and began deploying a fix at 14:53 UTC. Full service stabilization was only achieved at 16:44 UTC — nearly two and a half hours after the incident began.

Just a few hours later, at 18:24 UTC, another incident occurred — this time specifically on the Claude.ai web interface. It was resolved within eight minutes, but for users, it meant a second outage within a single day.

As reported by NJTODAY and confirmed by The Verge, the outage brought work to a halt for thousands of developers, students, and enterprise clients. Users reported failed chat requests, 500-level errors on the API, and complete workflow interruptions in Claude Code and Claude Cowork.

Not a First — June Is a Critical Month for Anthropic

Claude's status page reveals a troubling pattern. Over the past three weeks, Anthropic has recorded dozens of incidents — virtually every day in June has seen some form of outage or elevated error rate. The most frequently affected are the latest Opus 4.8 models, which suffer repeated outages, sometimes several times a day.

Let's look at the numbers: over the last 90 days, Claude.ai shows an uptime of 99.12%, Claude Code 99.28%, and Claude API 99.41%. For comparison, enterprise services typically guarantee 99.9% availability (SLA). A 0.6 percentage point difference means over four extra hours of outages per month compared to standard expectations.

The Verge's assessment is telling: editor Richard Lawler noted of the outage that "its own status dashboard shows this is a pretty common occurrence." For a company approaching a valuation of one trillion dollars, that's not a good look.

Today's Status: Another Incident in Progress

The problems didn't end with yesterday's outage. Today, June 24, 2026, since 13:16 UTC, another incident is in "Investigating" status — elevated error rates on Claude Opus 4.8. Neither the cause nor an estimated fix time has been announced yet.

On top of this, since June 13, access to Claude Mythos 5 and Claude Fable 5 — Anthropic's two most powerful models — remains suspended. The reason was a US government security directive banning foreign nationals from accessing these models after Mythos breached National Security Agency (NSA) systems during an authorized test. Anthropic responded by disabling them globally, stating it cannot technically enforce nationality-based restrictions.

Why Is This Happening? Infrastructure Can't Keep Up With Growth

Anthropic is experiencing explosive growth. The company is currently the world's most valuable AI startup with a valuation of around one trillion dollars and confidentially filed for an IPO on June 1. Both the user base and infrastructure demand are exploding — especially with the rise of agentic workflows like Claude Code, which consume many times more computing power than regular chat.

The company is trying to diversify its hardware sources. According to published information, it is in talks to use Microsoft Maia 200 chips and already works with Amazon Web Services, Google Cloud, SpaceX (renting capacity at the Colossus 2 data center), Broadcom, and CoreWeave. SpaceX compute capacity alone reportedly costs Anthropic $150 million per month.

Agentic assistants like Claude Code place far higher demands on infrastructure than classic chat interfaces. Each request can involve dozens or hundreds of internal model calls, file reads, command executions, and code analysis. Multiply this load by hundreds of thousands of simultaneously working developers, and it's no surprise that servers occasionally "buckle under the pressure."

What It Means for Developers and Businesses

For developers and companies that rely on Claude, this series of outages is a warning signal. Relying on a single AI provider is risky — especially if your workflow depends on Claude Code or the API. Experts recommend having a backup strategy: alternative models (GPT-5.5, Gemini 3.1), local LLMs, or at least a plan to survive a multi-hour outage without losing productivity.

Some companies are already implementing multi-model architectures that automatically switch to another provider in the event of an outage. Claude may offer top-tier quality in coding and analysis, but if you can't count on it during critical moments, its added value diminishes.

Broader Context: AI Industry's Infrastructure Crisis

Anthropic is not alone. The entire industry is grappling with scaling challenges. OpenAI has faced its own ChatGPT outages in recent months, leading to restrictions on new user registrations. GitHub Copilot had to pause new signups due to infrastructure overload from agentic workflows. Microsoft admitted that AI costs for its own developers exceeded the cost of human employees.

The entire industry is hitting physical limits: data centers are being built more slowly than demand grows, Nvidia chips are allocated on quotas, and the energy intensity of AI infrastructure is so enormous that the UN warns AI could consume 3% of the world's electricity by 2030.

For Anthropic, the current situation is particularly sensitive — the company is on the verge of an IPO, and every outage that The Verge writes about undermines investor confidence. At the same time, following the Mythos and Fable 5 affair, it faces regulatory pressure from the US government, complicating operation of its most advanced models.