Anthropic's NSA Work Raises New Questions About AI-Driven Cyber Risk
Anthropic is deploying its Mythos artificial intelligence model for US National Security Agency offensive cyber operations, according to reporting by the Financial Times. The company has assigned about half a dozen engineers to support the arrangement, working as forward-deployed staff to tailor the model for specific applications.
The technology may be used to infiltrate networks operated by countries including China and Iran, according to people familiar with the matter. An Anthropic source told the Financial Times that "the best way to build a good defence is to build a good attack," and that geopolitical rivals are likely developing similar AI-enabled capabilities.
A Shift in Cyber Risk Assessment
The deployment highlights a broader concern across the insurance industry: advanced AI systems can automate attacks and identify software vulnerabilities at scale. Insurers and reinsurers are reassessing how AI tools may affect cyber threat frequency and loss severity.
Industry specialists have warned that increasingly autonomous AI systems could accelerate attacks, exploit vulnerabilities more quickly, and create new underwriting challenges. Many carriers are now revisiting their underwriting assumptions and risk models to account for AI-driven threats.
Accumulation Risk Takes Center Stage
Anthropic expanded access to Mythos this week to 150 organizations across 15 countries, broadening availability beyond its initial limited US rollout. The wider distribution has intensified industry discussions around cyber accumulation risk.
Insurance professionals have long warned that a single cyber event, compromised vendor, or software vulnerability can affect thousands of organizations simultaneously. Such incidents create correlated losses across multiple policyholders and challenge traditional catastrophe modeling and portfolio diversification strategies.
Advanced AI tools that can identify and exploit vulnerabilities across many systems at once may amplify this risk. A coordinated attack using AI could generate losses that test reinsurance protections and conventional approaches to cyber risk management.
The Anthropic-Pentagon Dispute
The NSA arrangement comes as Anthropic continues a dispute with the US Department of Defense over the use of its Claude models. Anthropic previously sought restrictions on government use of its technology, objecting to applications involving mass surveillance of US citizens and lethal autonomous drones.
The Pentagon designated Anthropic a "supply-chain risk" in response. Anthropic has challenged that designation in court. Both Anthropic and the Defense Department declined to comment on the reported NSA arrangement.
Anthropic filed for an initial public offering that could value the company at more than $1 trillion.
For insurance professionals, the convergence of advanced AI capabilities and cyber risk requires attention to how AI for Insurance applications may reshape underwriting models and portfolio management strategies. Understanding Generative AI and LLM capabilities becomes essential for assessing emerging threats to insured organizations.
Your membership also unlocks:
