Claude outage on June 23: paid users lost access, while Anthropic's government AI continued to operate
A 85-minute outage of Claude AI on June 23, 2026, left paid subscribers without access to key Anthropic services, while the specialized version for government agencies continued to function without interruption. This incident clearly demonstrated how deeply the company separates its public and federal infrastructure.
Users noticed that during widespread errors across various Anthropic products—from claude.ai to the API and Claude Code—the service for government organizations remained operational. A full timeline of events is available on the company's status page.
Timeline and Scope of the Outage
Anthropic specialists promptly launched an investigation and reported finding a technical solution within 35 minutes. However, overall disruptions persisted for 85 minutes, and the incident status was changed to "resolved" only two hours after the first signal was recorded.
The errors that occurred paralyzed several key platforms simultaneously:
- claude.ai
- Claude API
- Claude Code
- Console
- Cowork
Notably, the specialized Claude service for government agencies was not included in the official list of affected segments.
Two Levels of Access: Deliberate Separation
Such a deep separation of AI systems was intentionally implemented by the developers. The government version of Claude operates entirely independently from the commercial infrastructure. The platform is deployed on the specialized federal cloud service Palantir, which holds strict FedRAMP High certification.
Last year, the General Services Administration (GSA) provided a similar level of access to all three branches of government for just $1. Fully isolated environments are considered the classic standard for strictly regulated cloud services. This is why government AI systems continued to operate without interruption while consumer products faced large-scale outages.
The incident was only part of a very challenging period for the company's public segment. According to Anthropic's official status page, technical specialists recorded more than 20 incidents between June 9 and June 23. Users regularly encountered errors or complete system unavailability, which primarily affected the new flagship Opus 4.8.
Reaction and Conclusions
To effectively cover the rapidly growing demand, the company promptly reserved up to 5 GW of new computing capacity from Amazon. Additionally, management signed new lease agreements for modern data centers. Against this backdrop, the isolated government segment remained safe from the outset, as it was not part of this overloaded resource pool.
Expert opinion: This incident is a clear signal for the market. Paying commercial users are essentially subsidizing the development of AI that is primarily protected for government agencies. Amid the multiple-fold increase in load on public services, Anthropic must either radically rethink its fault tolerance architecture or prepare for a wave of discontent from key clients.