Anthropic's recent release of Claude Opus 4.7 introduced enhanced safeguards aimed at preventing misuse. However, these stricter measures have inadvertently hindered legitimate developer use, sparking significant complaints.
This move follows Anthropic's announcement of Mythos, a model deemed too powerful for public release in vulnerability discovery and exploitation. While this assessment might serve the company's own narrative on risk, Anthropic decided to use Opus 4.7 as a testing ground for heightened guardrails. The company stated, "We are releasing Opus 4.7 with safeguards that automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses. What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models."
Developers have voiced strong objections in Claude Code's GitHub repository. Complaints against Anthropic's Acceptable Use Policy (AUP) classifier have surged, making it difficult for users to complete legitimate work.
Increased security has led to a rise in false positives, with Claude becoming overly cautious and refusing harmless requests. A review of GitHub issues confirms a clear upward trend in AUP-related refusals.
Concerns about invalid refusals have been raised for months. From July through September 2025, there were typically two to three complaints per month, such as Issue #4373, "Memory authorization code from claude.ai triggers API policy error."
This rate increased to around five to seven complaints monthly between October and November 2025, exemplified by Issue #8784, "Claude 4.5 Throws API Error: Claude Code is unable to respond to this request for normal requests randomly."
December saw fewer complaints, potentially due to holiday slowdowns. However, by January, the number returned to approximately eight. Developer #16129, reporting "Repeated False AUP Violations in Claude Code," noted, "Technical software development conversations should not trigger AUP violations. The safety filter appears overly aggressive on benign content." Similar numbers were observed in February and March.
April marked a significant escalation, with developers filing over 30 reports of claimed false positives across security, general development, and scientific use cases. One notable example is Issue #48442, "Persistent AUP false positives — 40+ per 4 sessions, across unrelated projects (psychology book, web app, infra, bot)."