Yann LeCun rejects AI extinction panic, while Cohere warns cyber risk is the real emergency

Yann LeCun has dismissed claims that AI could kill everyone within three years, while other technology leaders argue the immediate danger lies in cyberattacks and misuse. The disagreement matters because the AI safety debate is not one argument, but several wearing the same oversized coat.

Yann LeCun Watch analysis

What happened

CNN reports that LeCun, the former Meta chief AI scientist, told a user there was “not a chance in hell” that AI would cause human extinction within three years. He argued that people who work mainly with bits and tokens can underestimate how difficult it is for software to operate in the messy physical world.

LeCun’s view is part of a wider pushback against what CNN describes as “doomer” claims. NVIDIA chief executive Jensen Huang reportedly put the chance of AI ending humanity by 2030 at zero, while Cohere chief executive Aidan Gomez called extinction-focused talk misleading and scary.

Gomez’s position is less dismissive. He said AI presents a serious national-security risk through cyberattacks, particularly if powerful systems are used by states or malicious groups. He described misalignment, where an AI system behaves in ways its operators did not intend, as a real concern but one the industry is becoming better at measuring and detecting.

Rumman Chowdhury, chief executive of Human Intelligence PBC and a former US science envoy for AI, also questioned the mechanics behind some extinction scenarios. She argued that concerns about hypothetical future catastrophes can distract from present harms including deepfakes, mental-health problems and job losses.

Why it matters

The useful distinction here is between probability, mechanism and consequence. Saying that extinction is unlikely is not the same as saying AI is safe, and identifying a plausible cyber threat does not prove that a catastrophic attack is imminent.

For readers, the practical takeaway is to ask what a warning actually describes. Is it about a current capability, a foreseeable misuse, or a chain of future events that has not yet been demonstrated? Those categories should not be bundled together simply because “AI risk” fits neatly on a conference badge.

The argument also exposes a policy split. If the main danger is engineering failure, companies may favour testing, monitoring and voluntary restraint. If the danger is large-scale misuse or incentives that reward speed, critics will argue that outside oversight and enforceable rules are needed.

Our read

LeCun is right to resist turning every alarming AI incident into evidence of impending extinction. But the sceptical case is strongest when it addresses specific mechanisms rather than waving away the entire category of risk.

Gomez’s cyber warning is more immediately actionable. Security teams, regulators and AI companies can test whether systems enable attacks, leak sensitive information or make offensive operations cheaper. That is less cinematic than a machine uprising, but considerably more useful to people trying to decide what to fix this quarter.

What to watch

• Whether AI companies publish comparable evaluations for cyber misuse and autonomous behaviour.
• Whether regulators focus on measurable present harms rather than only distant scenarios.
• Whether LeCun, Huang and other sceptics offer concrete tests that could prove their confidence misplaced.
• Whether independent researchers can reproduce claims that current systems are becoming easier to monitor and control.

Discussion spark: Should AI policy prioritise measurable present threats such as cyberattacks and deepfakes, or preserve substantial resources for low-probability extinction scenarios that may be harder to test?

Independent WittyWires tracker for public updates about Yann LeCun. Not affiliated with or endorsed by Yann LeCun; this is not an official account.