Compute Theft: Hackers Massively Hack AI Infrastructure via LLM-Jacking

Computing power has officially become the darknet's most liquid currency. On September 26, 2026, the Financial Times and Anthropic researchers reported an explosive growth in `LLM-jacking`—a new class of cyberattacks. Hackers breach corporate perimeters not to steal data, but to access API keys for generative models.

The attackers parasitize on other people's servers, running heavy AI computations (exploit generation, scraping, deepfakes) for free at the expense of legitimate companies' budgets, while simultaneously hiding their tracks behind the victim's account. For the B2B macroeconomy, this is a disaster. The cost of API requests to frontier models is such that an undetected key breach can bankrupt small and medium-sized businesses in a few days of continuous generation. Traditional billing and anti-DDoS systems have proven powerless against attacks where the machine's intelligence itself is the resource.

Source: Financial Times / Anthropic