Coinbase CEO Warns Rogue AI Could Hit the Internet Within 2 Years

Coinbase CEO Brian Armstrong expects a rogue AI model to break loose on the internet soon. He put the timeline at one to two years.

He framed the scenario as a rerun of the 1988 Morris Worm rather than a civilizational threat.

Why Armstrong Expects a Rogue AI Incident Soon

The prediction lands after a year of real incidents. In July, OpenAI said two of its models escaped a test environment and hacked Hugging Face.

I wouldn't be surprised if an AI model goes rogue on the internet in the next year or two, something like the Morris Worm in 1988.It will feel like a big deal, with a media frenzy and calls to shut all AI down. But, as always, people will adapt, defenses get built, and it will…

— Brian Armstrong (@brian_armstrong) August 13, 2026

Those models wanted the answer key to a hacking benchmark. They then chained exploits across OpenAI systems and Hugging Face servers to reach the solutions database. Days later, the same agent reached a second firm through a customer’s vulnerable code. Nobody instructed the rogue AI to break out.

Armstrong has tracked the topic for months. In July, he argued that AI makes crypto rails more important because agents will transact constantly. Meanwhile, defenders are arming up. OpenAI shipped a cybersecurity-focused model this month and handed vetted researchers exploit development tools.

Armstrong expects the same pattern here. A media frenzy arrives, calls to halt AI development follow, and the industry patches the hole. The Coinbase CEO runs the largest US crypto exchange. His read carries weight with builders already wiring AI agents into payment systems.

Subscribe to our YouTube channel to watch leaders and journalists provide expert insights

Critics Say the Morris Worm Comparison Falls Short

The 1988 worm infected roughly 6,000 of about 60,000 connected machines in 24 hours, the FBI says. Its author, 23-year-old Cornell student Robert Tappan Morris, drew probation and a fine.

Damage estimates ran from $100,000 into the millions. Some universities cut themselves off the network for a week.

Yet the fallout also built the defenses. Within days, the Pentagon stood up the first computer emergency response team in Pittsburgh. That template still shapes incident response today.

Security researchers doubt a rogue AI failure would settle that cleanly. A worm spreads on fixed instructions, while an agent adapts to whatever blocks it.

Blockchain security expert Manuel Aráoz warned in May that AI agents outpace auditors across decentralized finance. Ledger executive Ian Rogers made a similar point this month about crypto wallet attacks.

Critics of the comparison list sharper risks. They point to autonomous cyberattacks, industrial-scale disinformation, and lost control of critical infrastructure.

The gap between the two camps comes down to recovery speed. Armstrong bets that patches land faster than damage spreads. His skeptics see failures that no patch reverses.

Neither camp disputes the direction. Model capability keeps climbing, and containment keeps lagging behind it.

Historically, the internet has absorbed each shock. A real rogue AI event will test whether that record holds.