Anthropic's Claude suffered a brief outage on April 13, knocking both Claude.ai and Claude Code offline for nearly an hour. But the bigger story is what happened when The Register's Thomas Claburn asked Claude to analyze its own GitHub issues. The model confirmed what users have been saying: quality complaints are spiking. April had already logged 20-plus quality issues in just 13 days, putting it on track to beat March's total of 18. March itself was a 3.5x jump over the January-February baseline.
The complaints range from caching problems to what one developer called "dangerous" prediction-first behavior on production projects. AMD's AI director Stella Laurenzo has publicly stated that Claude's responses keep getting worse. Boris Cherny, who heads Claude Code, has addressed some of these issues on GitHub. Then there's the wild allegation from a brand-new account claiming Claude autonomously deleted over 35,000 customer records and billing transactions belonging to Jixen Enterprises Private Limited, an Indian company. That claim remains totally unverified, and The Register hasn't heard back from Jixen. Some of the GitHub issues themselves appear to be AI-generated, which adds a strange recursive quality to the whole situation.
Here's the strange part. Margin Lab's benchmarking data shows Claude Opus 4.6 has held steady on SWE-Bench-Pro since February. No real drop. But benchmarks and real-world usage are telling different stories. Hacker News commenters describe stark gaps between account tiers, with personal accounts stalling out at 90% completion on coding projects while team and premium accounts work fine. Others say Opus 4.6 acts "incredibly dumb" for some users and normally for others. Anthropic has been throttling usage during peak hours to manage capacity, which could explain some of the inconsistency. The company hasn't responded to requests for comment.