In a striking turn of events, both Google and Anthropic have moved to restrict third-party access to their AI models, with OpenClaw, a popular OAuth client, at the center of the controversy. Google’s recent decision to limit accounts of its AI Ultra subscribers—who pay $249.99 a month—accessing its Gemini model through OpenClaw came without warning and has sparked outrage among users. Meanwhile, Anthropic updated its terms just two days earlier to explicitly prohibit OAuth token usage in third-party tools, claiming that such practices enable “token arbitrage,” where users extract more value from subscriptions than intended.
Token arbitrage has emerged as a critical concern for AI companies, as it disrupts the economics of flat-rate subscription models. Anthropic engineers have pointed out that third-party tools like OpenClaw generate “unusual traffic patterns” that lack the necessary telemetry to maintain the integrity of their offerings. This shift in policy is a clear signal that both companies are tightening their grip on how their technologies are utilized, potentially reshaping the landscape for third-party developers.
In a contrasting move, OpenAI has openly endorsed the use of its Codex subscriptions with third-party tools, positioning itself advantageously in the competitive AI market. This endorsement comes at a time when OpenClaw, which has garnered over 200,000 stars on GitHub since its rapid growth began in November 2025, is transitioning to a foundation structure with backing from OpenAI. Peter Steinberger, the founder of OpenClaw, announced his intention to join OpenAI just days before this controversy erupted, signaling a significant shift in the third-party AI ecosystem.
The backdrop of security concerns further complicates the situation. Recent reports have revealed that 21,639 instances of OpenClaw were exposed, raising alarms about vulnerabilities in the platform. In the span of just one week, five CVEs were patched, and a variant of the Vidar infostealer was found to have exfiltrated full configuration details, including gateway tokens. Additionally, a supply chain attack via the ClawHub skills marketplace, dubbed ClawHavoc, has intensified scrutiny on the platform’s security measures.
As both Google and Anthropic take a firmer stance against unauthorized access through third-party tools, the implications for developers and users alike remain significant. With OpenAI’s proactive support for third-party integrations, the competitive dynamics in the AI space are set to evolve, potentially leading to a bifurcation where only the most secure and compliant tools thrive. The developments underscore the ongoing struggle between accessibility and security in the rapidly changing world of AI technology.
AI Disclosure: This article was produced with AI assistance based on publicly available sources. Flash Intel editorial standards apply.
Source: Implicator.ai
