Anthropic restored full operations across its Claude platform on Friday after a service disruption that affected users in the United States, Europe and parts of Asia for several hours.

The outage began at 17:08 Central European Time and disrupted access to the Claude web interface, application programming interface, and developer-focused tools. During the incident, users frequently encountered HTTP 529 “overloaded” messages as well as internal server errors.

Service recovery took place in stages. Anthropic first brought back access to its Opus 4.6 model at 17:25 CET before gradually restoring additional services. According to the company, normal operations resumed at 20:27 CET.

Anthropic said the disruption was caused by an infrastructure problem rather than a security incident. The company stated that customer data was not at risk during the outage.

The interruption follows a series of recent reliability issues. Anthropic experienced a separate disruption lasting nearly six hours on June 2, while additional service problems were reported on June 1 and June 3. Despite those incidents, the company reports a 30-day platform availability rate of 99.3 percent.

The timing of the outage comes as Anthropic faces increased attention surrounding the security of its autonomous agents and advanced AI systems.

The company is currently examining reports of unauthorized access involving “Claude Mythos,” a cybersecurity tool that has not yet been publicly released. Anthropic said there is currently no evidence that its systems were compromised. Investigators are instead focusing on the possible misuse of access privileges within private forums.

Security researchers have also identified vulnerabilities affecting Anthropic’s development tools. RyotaK of GMO Flatt Security reported approximately 50 techniques capable of bypassing safeguards in Claude Code.

One vulnerability, assigned a CVSS 4.0 severity score of 7.8, allowed attackers to steal environment variables and credentials through a manipulated GitHub entry. Anthropic addressed the issue through security updates released in early June.

Additional concerns emerged on June 3 with the discovery of the “Miasma” worm, which targets AI-powered coding environments including Claude Code and VS Code. The malware spreads through malicious commits in GitHub repositories and can extract developer credentials when a repository is opened in an AI-assisted environment.

Reliability and security have become increasingly important as Anthropic expands the role of AI within its own software development processes. According to a company report published on June 5, Claude now generates the majority of code deployed into Anthropic’s production systems.

The report said more than 80 percent of code integrated into the company’s internal environment is generated by Claude. Anthropic engineers reportedly merged eight times more code per day during the second quarter of 2026 than they did in 2024.

The company also highlighted a recent achievement involving Claude Opus 4.8. On May 29, the model identified a four-year-old flaw in the Zcash cryptocurrency protocol that was linked to a market-value fluctuation estimated at €5 billion.

At the same time, Anthropic co-founder Jack Clark has raised concerns about the pace of AI development and the adequacy of existing safeguards. In a recent interview, he compared the industry to a vehicle equipped with an accelerator but no brakes, arguing for a verifiable global mechanism capable of pausing development when safety measures fail to keep pace with technological progress.

Meanwhile, Anthropic has expanded “Project Glasswing,” making its Mythos vulnerability-discovery tool available to 150 organizations across 15 countries, including the Australian government. During testing, the system identified more than 10,000 vulnerabilities classified as high or critical severity over roughly six weeks.