Will AI capability advances favor cyber attackers or defenders?

Bret Taylor · Oct 2, 2024 · hedged

For most of the problems created by AI there are AI solutions, and a white-hat/black-hat dynamic like the one in cybersecurity will give people the tools to judge whether information is authentic.

The same pattern played out in cybersecurity, and all the great AI minds are working on how this benefits humanity, so for every problem there will be an entrepreneur or researcher meeting the challenge.

Scope: framed as 'my hope' for the white-hat outcome

58:15 20VC: Bret Taylor: The AI Bubble and What Happens Now | How the Cost of Chips and Models Will Change in AI | Will Companies Build Their Own Software | Why Pre-Training is for Morons | Leaderships Lessons from Mark Zuckerberg

Brendan Foody · Jun 1, 2026

There will be an enormous boom in AI security engineering tools and AI-driven cyber defense

Attackers are already using swarms of coding agents to breach systems — as happened in Mercor's own incident — and the wave of AI-enabled cyber incidents is just getting started, so labs and enterprises are focused on building the best AI security engineer

8:59 20VC: Mercor CEO on Why Application Layer Companies Have No Defensibility, The Model is the Product | Token Spend Will Exceed Headcount Spend in 5 Years | The True Cost of Hiring AI Researchers in the Valley Today with Brendan Foody

Aaron Levie · Apr 20, 2026

The idea that early access to a frontier model would let you go secure and upgrade all your systems is an oversimplification, because upgrading software is a multi-year effort and offense/defense leapfrogging never ends

Unless a model were kept closed for a decade there is no magical moment where everything gets secured; security is an ongoing endless cycle

5:03 Endless leapfrogging no decisive defensive window

Anastasios Angelopoulos · Aug 3, 2026

We are about to see an insane wave of cyber attacks, and attackers are already trying to infiltrate American businesses at scale by applying for jobs as AI-generated fake people

At Arena, candidates pass full technical interviews and appear normal, then turn out not to exist — and other companies are seeing the same thing

33:08 Ai generated fake employees are already infiltrating companies

Brendan Foody · Jun 1, 2026

Swarms of coding agents make attackers far more dangerous because they can exhaustively review an entire codebase rather than being bound by human review speed and team size

A human attacker can only review a limited portion of code at human speed; agent swarms review everything — the whole codebase, the whole front end — and move much faster

9:48 Agent swarms give attackers exhaustive scale advantage