Security Experts: Anthropic's Fable Can Launch Autonomous Cyberattacks Within Months

As The Guardian first reported, national security experts have concluded that Anthropic's Fable model — and comparable frontier AI systems — are already capable of planning and executing sophisticated cyberattacks on governments and businesses, with deployment potentially months away. The assessment is not based on theoretical capability projections; sources cited in the report describe Fable as able to autonomously identify and exploit vulnerabilities at a scale and speed that surpasses existing offensive hacking tools.

The concern centers on autonomous operation: Fable does not need a skilled human operator to run an attack campaign. It can identify targets, chain exploits, and adapt to defenses without human intervention — a capability profile that moves AI from a hacking aid into a hacking actor in its own right. Critical infrastructure, financial systems, and government networks are the scenarios most frequently cited by the experts interviewed.

The report is already driving renewed pressure for mandatory pre-deployment safety evaluations and international coordination on frontier AI systems. Anthropic has not publicly disputed the capability claims. The company has built its brand around safety-first development, which makes the convergence of those two facts — safety-first lab, weapons-capable model — an uncomfortable thing to sit with.

Read the full story at The Guardian