Claude elevated errors have plagued Anthropic’s flagship models since late March 2026, with recurring service disruptions affecting both Sonnet 4.6 and Opus 4.6 across multiple product surfaces. The reliability crisis coincides with a catastrophic source code leak on March 31 that exposed the entire Claude Code repository, raising serious questions about Anthropic’s operational stability as enterprise adoption accelerates.
Key Takeaways
- Claude Sonnet 4.6 and Opus 4.6 experienced elevated error rates on April 4, March 26, and March 23, 2026.
- Anthropic accidentally shipped Claude Code source code (512,000 lines, 1,906 TypeScript files) to npm on March 31.
- Leaked code contained 44 hidden feature flags, including a Tamagotchi pet feature not publicly documented.
- Azure AI Foundry deployment of Opus 4.6 shows “No available regions” errors despite supported regions.
- GitHub repositories of the leaked code reached 50,000 stars and 41,500+ forks within hours.
The Timeline: Multiple Outages in Two Weeks
Claude elevated errors are not a new problem. On March 23, 2026, Claude.ai users experienced elevated errors between 9:10 PT and 9:26 PT, a brief but disruptive window. Three days later, on March 26, Anthropic began investigating elevated errors on both Opus 4.6 and Sonnet 4.6, with the initial alert flagged at 21:56 UTC and investigation starting at 23:06 UTC. On March 31, Opus 4.6 hit elevated error rates again, this time resolved by 09:44 UTC. The pattern accelerated into April: on April 3, Sonnet 4.6 experienced elevated errors that Anthropic resolved by 19:21 UTC, followed by reports of degraded response quality described as “brain fog” on Claude Opus 4.6 via Claude Code. Most seriously, on April 4, both Sonnet 4.6 and Opus 4.6 experienced elevated error rates across most product surfaces between 10:09 and 10:31 PDT.
This is not a single incident—it is a pattern. For a company positioning Claude as the enterprise alternative to GPT-4, recurring outages every few days undermine the core value proposition of reliability.
The March 31 Source Code Disaster
The reliability issues are compounded by a separate catastrophe that exposed Anthropic’s internal development practices in embarrassing detail. On March 31, 2026, Anthropic accidentally shipped the entire Claude Code source code—version 2.1.88, totaling 59.8 MB with 512,000 lines of code across 1,906 TypeScript files—to the public npm registry due to a misconfigured debug file. The R2 bucket hosting the code was publicly accessible, and the leak was discovered at 04:23 UTC by researcher Chaofan Shou before being pulled around 08:00 UTC.
What makes this leak particularly damaging is not just the volume of exposed code but what it revealed. The leaked source contained 44 hidden feature flags, including a Tamagotchi pet feature that was never publicly announced. This suggests Anthropic has experimental functionality baked into production code that users are unaware of—a red flag for both security and product roadmap transparency. Anthropic’s official response dismissed the incident as “human error, not a security breach,” but the distinction feels semantic when the company’s entire development pipeline is accessible to the public.
Aftermath: Forks, DMCA Requests, and Decentralized Mirrors
The internet moved fast. GitHub repositories containing the leaked code hit 50,000 stars in under two hours and accumulated 41,500+ forks, making it virtually impossible to fully contain. Anthropic issued DMCA requests, but by then the code had already been mirrored on decentralized platforms like Gitlawb and rewritten in Python via clean-room implementations. This marks Anthropic’s second security slip-up in days, compromising roughly 1,900 files and exposing the company’s development practices to competitors and security researchers worldwide.
The leak also revealed deployment friction. On April 3, 2026, users attempting to deploy Claude Opus 4.6 in Azure AI Foundry encountered “No available regions” errors despite regions being listed as supported. Anthropic attributed this to a phased rollout with pending backend entitlements and capacity constraints, suggesting the company is struggling to scale infrastructure alongside growing demand.
What This Means for Claude Users
For developers and enterprises relying on Claude via API, the message is clear: Anthropic is experiencing both operational and security growing pains. Recurring elevated errors across multiple models indicate systemic issues, not isolated incidents. The source code leak reveals that production code contains undocumented experimental features, raising questions about what else might be hidden in Claude’s behavior. And deployment bottlenecks on Azure suggest Anthropic’s infrastructure cannot keep pace with adoption.
This does not mean Claude is fundamentally broken—the outages are being resolved within hours, and the models still perform well during normal operation. But the frequency and breadth of these issues suggest Anthropic prioritized speed-to-market over operational maturity. For users evaluating Claude against alternatives, reliability and transparency matter as much as raw capability.
Is Claude down right now?
As of April 6, 2026, no incidents are reported on Claude’s status page. However, given the pattern of outages every few days in late March and early April, checking status.claude.com directly before deploying critical workloads is prudent.
What was exposed in the Claude Code leak?
The March 31 leak exposed 512,000 lines of Claude Code source code, including 44 hidden feature flags such as a Tamagotchi pet feature not publicly documented. The entire development pipeline became visible to the public, and despite DMCA requests, the code was quickly mirrored on decentralized platforms.
Why is Claude experiencing degraded performance?
Anthropic has not provided a detailed root cause analysis for the recurring elevated errors. Reports of “brain fog” on Opus 4.6 via Claude Code suggest quality degradation, possibly linked to infrastructure scaling issues or model deployment problems during the phased Azure rollout.
Anthropic faces a credibility test. The company can address elevated errors through infrastructure investment and incident response improvements, but rebuilding trust after exposing 512,000 lines of source code and hiding features from users will take longer. For now, Claude remains powerful but unreliable—a risky foundation for mission-critical applications.
Edited by the All Things Geek team.
Source: TechRadar
