Did Anthropic Just Cross the “Don’t Be Evil” Line?

Detailed Analysis

A Reddit post titled "Did Anthropic Just Cross the 'Don't Be Evil' Line?" captures a growing wave of user disillusionment with the company, cataloging a series of grievances that together paint a picture of eroding trust. The author, a self-described longtime Anthropic supporter, argues that the company's trajectory since the Opus 4.6 release has shifted from principled AI safety leadership toward cost-cutting decisions dressed up in safety rhetoric. The complaints span several distinct issues: perceived quality degradation in newer models (Sonnet 5 and Opus 5), the controversial Mythos/Fable release cycle that combined dire safety warnings with what the author calls "stupidly simple" protections, silent model downgrades that left users unable to verify which model version they were actually interacting with, and reduced context windows even for previously trusted models like Opus 4.6 within Claude Code.

The specificity of these complaints matters because Anthropic has built its entire brand identity around being the "responsible" AI lab—the company that talks openly about existential risk, publishes extensive safety research, and positions itself as more cautious than OpenAI, Google, or Meta. This reputation has been central to its enterprise sales pitch, its recruiting of safety-focused researchers, and its public narrative during congressional testimony and media appearances. When users perceive a gap between stated values and actual behavior—particularly around transparency about model versions and capabilities—it strikes at the core of that brand promise. The reference to Google's abandoned "Don't be evil" motto is a pointed comparison, invoking a well-known cautionary tale about how mission-driven tech companies can drift toward extractive, opaque practices once market pressures and monetization needs intensify.

The specific grievance about silent downgrades is particularly significant in the context of ongoing industry debates about model transparency. Users and developers increasingly rely on consistent, predictable model behavior for building products, workflows, and automations. When a company changes model performance, context windows, or underlying weights without clear disclosure, it undermines the reliability that professional users depend on—especially for a company like Anthropic that markets Claude heavily toward developers and enterprises via Claude Code and the API. The accusation that Anthropic simultaneously criticizes open-source competitors as "dangerous" while allegedly cutting corners on its own service quality also touches a live nerve in AI discourse: critics of closed AI labs have long argued that safety rhetoric can function as a competitive moat, discouraging open alternatives while consolidating power and pricing leverage in the hands of a few well-funded firms.

More broadly, this post reflects a maturing phase in the AI industry where the honeymoon period of unconditional trust in "mission-driven" labs is giving way to more skeptical, consumer-style scrutiny. As Anthropic reportedly moves toward a future IPO and faces intensifying competitive and margin pressures from rivals like OpenAI, Google DeepMind, and a fast-growing open-weight ecosystem, the tension between commercial incentives and stated ethical commitments becomes harder to paper over. The article referenced within the post—a New York Times opinion piece questioning whether AI danger narratives from labs like OpenAI and Anthropic serve strategic marketing purposes—suggests this skepticism is not confined to Reddit but is entering mainstream discourse. For Anthropic, sustaining its differentiated reputation as the safety-conscious lab will likely require more transparent communication about model changes, pricing tradeoffs, and capacity constraints, particularly as its user base grows more technically sophisticated and quicker to notice discrepancies between marketing claims and lived product experience.