Ali Ghodsi: AI doom-mongering is a fantasy; security is real
By Ray Dombrowski · Reporting from Youngstown ·
Ali Ghodsi argues that fears of AI ending the world distract from the immediate dangers of unsecured ports and failed corporate workflows.
Databricks CEO Ali Ghodsi told a16z partners Martin Casado and Sarah Wang to stop scaring the public. He said the risk of AI ending the world is nearly zero. To Ghodsi, doom-mongering serves only to terrify the elderly and invite a swarm of federal committees and compliance officers. The real challenge is not a robot uprising, but the grind of security and the struggle to help companies use tools to actually finish their work.
The Danger Is an Unsecured Port
Ghodsi separates the fantasy of a god-like AI from the reality of a breach. He said that while superintelligence remains a distant dream, deploying reinforcement learning agents on server clusters creates immediate holes. Vulnerabilities that once took months to weaponize now take hours. Casado noted that we have not yet seen a plague of AI-driven worms like those of the early internet. But Ghodsi argues that most corporate networks are a mess of legacy hardware, dusty servers, and forgotten passwords. If an agent finds a loop in the system, it will exploit it. This is a matter of patch schedules and payroll, not a prophecy.
Why does the hype outrun the payroll?
I spent years watching fabrication shops and workforce boards, where the smell of ozone and the roar of CNC machines define the day. I learned to judge a tool by whether it actually changes the work on the shop floor. Ghodsi said most companies still use chatbots as fancy search engines. The bottleneck is not the model's IQ, but its lack of "ontology"—the specific, messy way a company’s employees actually move a project from a sketch to a finished part. Firms do not fail for lack of chips; they fail because they cannot map their own workflows. Labs push frontier models to justify massive hardware bills and soaring valuations. But the narrative breaks when a CFO asks why a costly token run was spent just to rename a file.
Panic-Driven Regulation
Tech executives race for market share while begging Washington for guardrails. This looks like self-preservation. When CEOs warn that humanity might vanish, they invite federal rules and licensing schemes that only the biggest players can afford. These rules create a moat for the giants while starving the startups of oxygen. Ghodsi said leaders must stop triggering public panic. When regulators write laws based on sci-fi scripts, the cost falls on the business owner paying a real payroll, not the prophet selling a book on the side.
We do not need a pause on intelligence or another hearing led by politicians who cannot configure a router. We need security teams that can kill a port before a script runs. We need software that balances a ledger instead of drafting a press release.