Anthropic’s run-in with the Pentagon over Claude guardrails and a China-based distillation campaign against its IP highlight the external pressures that top AI models can pass to enterprises downstream.

CSO Online offers insights into cybersecurity, risk management, and IT leadership, providing articles, reports, and expert analysis to help security professionals navigate the evolving threat landscape and protect their organizations from cyber attacks. By exploring CSO Online's curated content, CISOs, security managers, and IT executives can learn about threat intelligence, security frameworks, and incident response strategies for building resilient cybersecurity programs. Whether you're defending against ransomware, phishing attacks, or insider threats, CSO Online offers resources to strengthen your security posture and safeguard your digital assets.

CSO Online

Two high-profile incidents involving Anthropic's Claude reveal a new category of enterprise risk: frontier AI models are now subject to external pressures that CISOs must factor into deployment decisions. A China-based extraction campaign by DeepSeek, Moonshot AI, and MiniMax ran over 16 million interactions through ~24,000 fraudulent accounts to map Claude's agentic reasoning and coding capabilities. Simultaneously, the Trump administration banned Claude from federal use after Anthropic refused to remove guardrails around autonomous weapons and mass surveillance. When Anthropic declined, OpenAI and xAI quickly moved to fill the gap. The core risk for CISOs: frontier models can be profiled, studied, or pressured upstream before reaching enterprise environments, and those forces shape how the model behaves once deployed. The recommendation is to treat frontier AI as a high-value dependency exposed to upstream influence and build monitoring to detect when those forces manifest internally.

Anthropic AI ultimatums and IP theft: The unspoken risk