Anthropic’s AI Paradox: NSA Cyber Operations and a Global Pause Call
Anthropic, a prominent developer in the artificial intelligence landscape, finds itself at the center of a compelling paradox concerning the advancement of cutting-edge technology. On one hand, the company is reportedly collaborating with government entities like the National Security Agency (NSA) to deploy its powerful AI models for offensive cyber operations. On the other, Anthropic publicly advocates for a global moratorium on further AI development, warning of the existential risks associated with autonomous self-improving systems.
NSA Collaboration: Mythos in National Security
Recent reports from the Financial Times indicate that approximately six Anthropic engineers have been embedded within the NSA to help deploy Mythos—the company’s most capable AI model—for offensive cyber operations. These forward-deployed staff are customizing the model for specific applications, which sources suggest could be useful for infiltrating networks in countries like China and Iran. Mythos is the very model Anthropic has declined to release publicly, citing misuse risk, and has limited to vetted partners through Project Glasswing, a restricted coalition that includes Microsoft, Apple, and Amazon.
“This deployment highlights the complex geopolitical landscape where cutting-edge AI capabilities are becoming critical tools for national security agencies,” remarked a cybersecurity analyst.
Notably, this collaboration continued even amidst Anthropic‘s legal dispute with the Pentagon. In late February, Defense Secretary Pete Hegseth designated the company a ‘supply-chain risk’ after a $200 million contract collapsed. The sticking point was Anthropic‘s refusal to let the DoD use its Claude model for fully autonomous weapons or domestic mass surveillance. The NSA contract, however, was exempt from that ban.
The Call for a Global Pause: Concerns Over Self-Improving AI
On the same day the NSA story broke, Anthropic‘s internal research institute published “When AI Builds Itself,” a report detailing how far its Claude model has come in automating its own development. The report reveals that Claude now writes over 80% of the code merged into Anthropic‘s production codebase, a significant increase from early 2025. Engineers are now shipping roughly eight times as much code per day as they did in 2024.
The report’s authors, Anthropic Institute lead Marina Favaro and co-founder Jack Clark, argue this trajectory is heading towards what they term ‘recursive self-improvement.’ This refers to AI systems that autonomously design, build, and train their own successors, with human involvement diminishing at every step.
A striking data point cited is an experiment where Claude agents were tasked with an open AI safety problem—whether a weaker model can reliably supervise a stronger one. Over 800 cumulative compute hours, the agents recovered 97% of the performance gap between the models, compared to 23% achieved by two human researchers over about a week. This marks the first published instance of Claude exercising research judgment, rather than merely executing specified tasks.
“The concept of AI systems designing their own successors fundamentally shifts the paradigm of human control, presenting unprecedented challenges for governance and safety,” noted an AI ethics researcher.
The company proposes a verifiable global pause—multiple frontier labs halting simultaneously with independent verification—likening it to Cold War-era nuclear treaties. Anthropic states it would join such an initiative, acknowledging that a unilateral slowdown would only cede the lead to others.
Industry Contradictions and the Future of AI
This situation mirrors broader trends within the AI industry. In 2023, over a hundred prominent figures in the AI research community signed an open letter advocating for global efforts to mitigate the existential risks posed by AI development. Yet, despite these warnings, development has not ceased. OpenAI continued, and Anthropic continued. As AI remains one of the most profitable business ventures of the decade, even those sounding the alarm are reluctant to halt progress.
The Pentagon’s deadline to drop Claude from its systems falls in August, around the same time Anthropic‘s IPO is expected to bring its finances into public view. This timing underscores the tension between ethical considerations, national security imperatives, and the immense commercial interests shaping the future of artificial intelligence.
Frequently Asked Questions (FAQ)
-
What is the Anthropic Paradox?
The Anthropic Paradox refers to the company’s dual stance: simultaneously providing its advanced AI models for offensive cyber operations to government agencies (like the NSA) while actively advocating for a global moratorium on AI development due to potential existential risks.
-
Which AI model is the NSA using from Anthropic?
The NSA is reportedly using Mythos, which is Anthropic‘s most capable AI model and has not been released to the public due to concerns about misuse.
-
Why is Anthropic calling for a pause in AI development?
Anthropic is calling for a pause due to concerns about ‘recursive self-improvement’ in AI, where AI systems could autonomously design and train their successors, potentially leading to a loss of human control and unpredictable outcomes.
