How AI is rewriting the rules of cyber defence

This article summarizes an episode of Lightspeed Venture’s video series featuring Vitaly Gudanets, CISO of Anthropic.

Vitaly Gudanets, chief information security officer (CISO) at Anthropic, an AI research company, says that security experts are falling behind AI-powered attackers. This imbalance forces a change in how companies defend themselves and shows the weakness of old security methods.

Attackers have seized the advantage

The fast growth of AI has created an imbalance between attackers and defenders. Security experts are now playing catch-up against threats that are faster and larger than ever before.

Defenders are behind
Daily security news shows how quickly today’s threats are evolving. For people doing the work, the feeling of falling behind smarter attackers is real.

Gudanets says, “As a practitioner, I think we’re behind. You see it every day in the news. Attackers are getting smarter. They’re getting more efficient. Time to market for exploits is getting faster and faster, and we’re behind.”

A cautious culture creates problems
The traditional approach to system defence, which relies on people manually reviewing everything, has become a weakness against automated attacks.

Gudanets explains, “Some of it has to do with a lack of trust. It’s hard to tell what’s real and what’s not. As defenders, we’re taking a slower approach. We always want to have a human in the loop.”

Response times have disappeared
Security teams had time to identify, analyze, and fix problems. That time is gone, leaving them constantly reacting to new attacks.

Gudanets notes, “The need to detect attacks is now becoming a matter of minutes, if not seconds. In the past, we had time to detect and respond. Now it happens so quickly, we get overwhelmed.”

AI is the only countermeasure

This shift requires a new approach. Defenders need to use AI tools to counter AI attacks and work more effectively.

Security teams must fight AI with AI
To get ahead of attackers, companies need new tools that go beyond manual defence. AI can help by predicting threats and automating complex tasks that move too fast for humans to handle.

Gudanets explains, “You could see the value coming to the market using the models. An automated SOC, predictive detection capabilities, and automated red teaming are possible. Models are really good at finding vulnerabilities.”

Anthropic models improve code security by default
Gudanets says Anthropic relies on AI as part of its software development process. “Internally within Anthropic, we rely heavily on Claude in everything we do. Basic classes of vulnerabilities like cross-site scripting and SQL injection are starting to get eliminated in the codebase.”

The job is to enable, not block

Success depends on speaking the language of business

AI models are national security assets

Hiring for a high-pressure job

---