BCG warns AI‑driven cyber attacks outpace defensive adoption

BCG’s report crystallizes an uncomfortable reality: offensive use of AI in cyberspace is compounding faster than defensive deployment. Attackers now routinely use generative models for phishing, deepfake‑enabled fraud, and automated vulnerability discovery, while most enterprises still rely on slow, manual playbooks. That means the digital substrate on which AI labs, data centers, and model supply chains sit is becoming more fragile just as we lean on it to train and deploy ever larger systems. ([wvnews.com](https://www.wvnews.com/news/around_the_web/partners/pr_newswire/subject/surveys_polls_and_research/ai-driven-cyber-threats-are-outpacing-defense-capabilities/article_efe5cdaf-ec65-5c3b-be6a-cff60eea09a9.html))

From an AGI‑race perspective, this widens the risk surface. Advanced labs are attractive targets for state and criminal actors seeking model weights, proprietary data, or leverage over critical infrastructure. If only a small minority of organizations are using AI for their own defense, then frontier labs and their cloud providers will carry disproportionate responsibility for hardening the ecosystem. Expect more emphasis on “autonomous SOCs” and AI‑in‑the‑loop security tooling as a result.

The strategic implication is that AI safety can’t be siloed to model alignment and red‑teaming. Resilience against AI‑enhanced cyberattacks is now part of the same conversation. If the world can’t keep core infrastructure secure as models get more agentic, societies may be forced to slow or constrain deployment regardless of what technical alignment progress looks like.