OpenAI and Anthropic pitch AI cyber-defence tools as AI-fuelled hacking surges
After a $6.3bn North Korean heist on Binance, frontier AI labs position cyber-security as a major new revenue line even as their models arm hackers.
What to know
- North Korean hackers stole $6.3bn from Binance on October 29th, the FBI says, after a $2bn haul last year that was triple the level two years earlier, per Chainalysis.
- OpenAI and Anthropic are both pitching cyber-security tools to enterprises; Anthropic's IPO filing calls it 'one of our most important near-term opportunities.'
- Industry figures predict cyber-security software spending, already $150bn a year, could grow dramatically; OpenAI's Greg Brockman says it may eventually 'vastly outstrip' spending on all other software.
- Independent testing shows frontier models are still weak at prioritising and fixing vulnerabilities once found: Anthropic's best model passed under half of a Collinear AI benchmark in September, though a Google model scored 71% in October.
“one of our most important near-term opportunities”
Anthropic, AI lab, IPO filing · The Economist ↗ · Oct 31
Sarah Friar OpenAI chief financial officer
Greg Brockman OpenAI co-founderAnthropic AI lab, OpenAI rivalNazneen Rajani CEO, Collinear AIBobby Kuzma Analyst, IDCNorth Korean hackers Alleged perpetrators of the Binance theft
How it unfolded 1 development · click the chart to see its coverage articlesposts
-
1
Economist details OpenAI and Anthropic's cyber-security push
The Economist reports that OpenAI's finance chief Sarah Friar expects cyber-security to become a huge business line, Anthropic's IPO filing names it a top near-term opportunity, and Goldman Sachs has raised the cyber-security share of its AI budget, even as experts warn the business case remains unproven.
“huge business…”
— Sarah Friar -
background
Hackers steal $6.3bn from Binance; FBI blames North Korea — Binance, the world's largest crypto exchange, was hit by a $6.3bn theft; Binance said attackers gained access via one of its engineers' laptops, and the FBI has pointed to North Korea.
-
background
Google model jumps to 71% on cyber benchmark; investors push spending — A new Google model scored 71% on the Collinear AI test, while Andreessen Horowitz urged portfolio companies to consider doubling their cyber-security outlays.
-
background
Collinear AI benchmark exposes models' weak vulnerability-fixing — A new cyber-security test from Collinear AI measured how well models could find and fix vulnerabilities; Anthropic's best model at the time passed fewer than half the tasks.
-
background
Anthropic withholds its Mythos model from the public — Anthropic kept an advanced model called Mythos out of public release, instead offering it to select software providers and financial institutions to find gaps in their cyber-defences; OpenAI soon launched a similar offering.
-
background
North Korean hackers steal $2bn in crypto, triple prior rate — Chainalysis data cited by the Economist shows North Korean cyber-thieves stole $2bn-worth of cryptocurrency last year, three times the level two years before, setting the stage for a bigger 2026 haul.
Also covered reported alongside — the timeline has no entry for these yet
-
first by Hindustan Times, 9d ago · also The Economist