Anthropic Tightens Claude Access Rules, Faces Developer Backlash Over OpenClaw Use

New updates have been reported about Anthropic.

Anthropic is moving to exert tighter commercial and technical control over how its Claude models are accessed via third-party agents, a shift that has sparked friction with the open-source OpenClaw community and raised questions about the company’s platform strategy. The company recently confirmed that standard Claude subscriptions will no longer cover use through external “claw” harnesses such as OpenClaw, forcing those workloads onto metered API billing and effectively introducing what developers have dubbed a “claw tax.”

Anthropic says the change is driven by the economics and infrastructure impact of claw-style agents, which often run continuous reasoning loops, repeatedly retry tasks, and integrate with multiple external tools, making them significantly more compute-intensive than typical prompt-based usage and misaligned with flat-rate subscription plans. At the same time, Anthropic is investing in its own agent product, Cowork, including features like Claude Dispatch for remote task assignment and control, intensifying perceptions among some developers that the company is prioritizing its closed ecosystem over open-source third-party tooling.

Tensions escalated when OpenClaw creator Peter Steinberger, now employed by rival OpenAI, reported that his Anthropic account was suspended for “suspicious” activity even though he said he had migrated his Claude usage to API-based billing in line with the new policy. Following public attention, Anthropic restored his account within hours, and an Anthropic engineer publicly stated that the company has never banned users for running OpenClaw and offered technical assistance, suggesting the suspension may have been an enforcement error or automated risk flag rather than a deliberate policy move.

Steinberger has nevertheless criticized Anthropic’s timing and motives, arguing that the company rolled out premium features in its own Cowork agent, then tightened pricing and access terms around competing open-source harnesses that enable multi-model usage. His remarks underscore a broader strategic tension: Anthropic must balance monetizing high-intensity agent workloads and protecting its product roadmap with maintaining goodwill among developers who value Claude for neutrality and interoperability.

For Anthropic’s business, the new API-only rule for claws should improve revenue alignment with actual compute consumption, potentially lifting average revenue per user and smoothing infrastructure planning, but it risks accelerating efforts by OpenAI and others to attract OpenClaw users who currently favor Claude. Developer sentiment will be a key indicator to watch, as sustained perceptions of legal pressure or account risk could discourage integration of Anthropic models into open-source agent frameworks, affecting future ecosystem depth and enterprise experimentation. While Anthropic has yet to fully clarify the circumstances around Steinberger’s suspension, its handling of similar edge cases and communication with third-party tool builders will influence how the market views Anthropic’s openness relative to competing AI platforms.