Mystery solved: Anthropic reveals changes to Claude's harnesses and operating instructions likely caused degradation

Mystery solved: Anthropic reveals changes to Claude's harnesses and operating instructions likely caused degradation

Mystery Solved: Anthropic Reveals Changes to Claude's Harnesses and Operating Instructions Likely Caused Degradation

Anthropic, the company behind the popular AI model Claude, has finally addressed the controversy surrounding the perceived degradation of its flagship model. For several weeks, a growing chorus of developers and AI power users reported a phenomenon they described as "AI shrinkflation," where Claude seemed less capable of sustained reasoning, more prone to hallucinations, and increasingly wasteful with tokens. Critics pointed to a measurable shift in behavior, alleging that the model had moved from a "research-first" approach to a lazier, "edit-first" style that could no longer be trusted for complex engineering.

Background & Context

The controversy gained momentum in early April 2026, fueled by detailed technical analyses from the developer community. Stella Laurenzo, a Senior Director in AMD's AI group, published an exhaustive audit of 6,852 Claude Code session files and over 234,000 tool calls on GitHub, showing performance falling from her usage before. Her findings suggested that Claude's reasoning depth had fallen sharply, leading to reasoning loops and a tendency to choose the "simplest fix" rather than the correct one. This anecdotal frustration was seemingly validated by third-party benchmarks, including a report by BridgeMind.

Impact on Swiss SMEs & Finance

The degradation of Claude's performance has significant implications for businesses and investors that rely on AI-powered tools for complex engineering and decision-making. The perceived shift from a "research-first" to a "edit-first" approach raises concerns about the accuracy and reliability of AI-generated solutions. Swiss SMEs, in particular, may be affected by the degradation, as they often rely on AI-powered tools to stay competitive in the market. The incident highlights the importance of transparency and accountability in the development and deployment of AI models.

What to Watch

As Anthropic has resolved the issues by reverting the reasoning effort change and the verbosity prompt, while fixing the caching bug in version v2.1.116, users can expect a return to the previous level of performance. However, the incident raises questions about the long-term reliability of AI models and the need for more robust testing and validation procedures. Readers should monitor Anthropic's future updates and developments to ensure that similar issues are addressed promptly and transparently.

Source

Original Article: Mystery solved: Anthropic reveals changes to Claude's harnesses and operating instructions likely caused degradation

Published: April 23, 2026

Author: carl.franzen@venturebeat.com (Carl Franzen)

---

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Consult a licensed financial advisor before making investment decisions.