# tech CEO's are fighting again! Source: https://www.youtube.com/watch?v=mfTAEQyC3qw Recap page: https://rapidrecap.app/video/mfTAEQyC3qw Generated: 2026-01-16T17:38:50.414+00:00 --- ## Quick Overview Anthropic reportedly cut off xAI's access to its Claude AI model, citing a breach of commercial terms of service, which xAI co-founder Tony Wu confirmed in an internal email, leading to significant backlash and subsequent lifting of all bans by Anthropic. **Key Points:** - Anthropic reportedly banned Elon Musk's xAI employees from using Claude, specifically Claude Code, due to violating commercial terms of service (00:09). - xAI co-founder Tony Wu acknowledged the block in an internal email, framing it as a push to develop xAI's own coding products (00:16). - The ban targeted third-party harnesses using Claude subscriptions, which Anthropic claimed generated unusual traffic patterns hindering debugging and support (00:58, 02:26). - OpenCode, a third-party tool, found a workaround to allow users to utilize Claude Code functionality via their own platform (01:04). - OpenAI was reportedly using Claude's API, particularly Claude Code, to benchmark and test GPT-5, which Anthropic's terms strictly prohibit for building competing systems (03:24). - Following community uproar, Anthropic's Tariq announced they lifted all bans and acknowledged their communication on the violation was unclear (02:54). - The video highlights that Anthropic's Opus 4.5 model achieved 80.9% accuracy on the SWE-bench Verified benchmark, outperforming competitors like GPT-5.1-Codex-Max at 77.9% (04:04). ![Screenshot at 00:10: An on-screen graphic illustrates the news headline: "Anthropic reportedly cut off xAI's access to Claude," showing images of Anthropic co-founder Dario Amodei and Elon Musk, symbolizing the conflict between the two AI labs.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-00-10.jpg) **Context:** The video discusses a recent escalation in the 'AI wars,' specifically between Anthropic (creators of the Claude models) and Elon Musk's xAI. The core conflict revolves around Anthropic cutting off access to its powerful Claude Code model for xAI employees, alleging that xAI was using the model via a third-party tool called Cursor or OpenCode to develop its own competing models, which violates Anthropic's commercial terms of service. The situation gained significant attention on social media, forcing a public response from both sides. ## Detailed Analysis The video reports on Anthropic cutting off access to its Claude AI models, specifically Claude Code, for employees of Elon Musk's xAI, citing a breach of commercial terms of service. This action was confirmed by xAI co-founder Tony Wu via an internal email, where he noted the block would hurt productivity but motivate xAI to develop its own coding products. The restriction specifically targeted third-party harnesses like Cursor/OpenCode that allowed users with Claude subscriptions to utilize the model outside of Anthropic's direct ecosystem. Anthropic argued this usage created debugging difficulties due to unusual traffic patterns and violated terms prohibiting the building or testing of competing systems. The situation was further complicated by reports that OpenAI itself was using Claude Code to benchmark GPT-5, a move also against the terms. Due to significant community backlash, Anthropic's Tariq later announced the lifting of all related bans, admitting their initial communication about the violation was unclear. The presenter also contrasted this competitive environment with performance benchmarks, showing Anthropic's Opus 4.5 achieving 80.9% accuracy on the SWE-bench, leading the field. ### Anthropic vs. xAI Conflict - Anthropic blocked xAI employees from using Claude Code, citing violation of commercial terms forbidding use for building competing AI systems - xAI co-founder Tony Wu confirmed the block in an email, framing it as motivation to develop internal tools (00:09-00:14, 00:16-00:28) ### Third-Party Tool Workaround - Tools like OpenCode used Claude Code functionality via a workaround, allowing users to continue using it for coding tasks outside the direct Claude interface, which Anthropic sought to prevent (01:04-01:39) ### OpenAI's Alleged Violation - OpenAI reportedly used Claude Code to benchmark and test GPT-5, which is explicitly prohibited by Anthropic's terms of service for developing competing models (03:24-03:32) ### Community Reaction and Resolution - Following public outcry, Anthropic's Tariq lifted all bans, apologizing for unclear communication regarding the Terms of Service violation (02:24-03:01) ### AI Model Performance Context - A chart shows Anthropic's Opus 4.5 leading the SWE-bench Verified (n=500) with 80.9% accuracy, while GPT-5.1-Codex-Max scored 77.9% (04:04-04:08) ### Anthropic's Ecosystem Strategy - Anthropic also announced 'Claude for Healthcare' and introduced a 'Cowork' feature to keep users within their ecosystem rather than relying on third-party wrappers (05:11-05:27) ![Screenshot at 00:10: A split screen graphic showing Anthropic's logo and personnel alongside Elon Musk, summarizing the core conflict between the two AI labs over Claude access.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-00-10.jpg) ![Screenshot at 00:17: A screenshot of an internal email from xAI co-founder Tony Wu confirming that Anthropic blocked all xAI employees from using Claude services, citing competitive restrictions.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-00-17.jpg) ![Screenshot at 00:56: A Hacker News post detailing the core issue: "Anthropic blocks third-party use of Claude Code subscriptions," highlighting the restriction on external tools.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-00-56.jpg) ![Screenshot at 03:12: A tweet from Dax showing that OpenAI was allegedly using Claude Code to benchmark and test GPT-5, which violates Anthropic's terms.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-03-12.jpg) ![Screenshot at 04:04: A bar chart titled "Software engineering SWE-bench Verified \(n=500\)" displaying Opus 4.5 achieving the highest accuracy at 80.9%, establishing the performance context of the LLMs involved.](https://ss.rapidrecap.app/screens/mfTAEQyC3qw/00-04-04.jpg)